< draft-floyd-tcp-highspeed-02.txt   draft-floyd-tcp-highspeed-03.txt >
Internet Engineering Task Force Sally Floyd Internet Engineering Task Force Sally Floyd
INTERNET DRAFT ICSI INTERNET-DRAFT ICSI
draft-floyd-tcp-highspeed-02.txt February, 2003 draft-floyd-tcp-highspeed-03.txt 29 June 2003
Expires: December 2003
HighSpeed TCP for Large Congestion Windows HighSpeed TCP for Large Congestion Windows
Status of this Memo Status of this Document
This document is an Internet-Draft and is in full conformance with This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026. all provisions of Section 10 of RFC2026.
Internet-Drafts are working documents of the Internet Engineering Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet- other groups may also distribute working documents as Internet-
Drafts. Drafts.
Internet-Drafts are draft documents valid for a maximum of six months Internet-Drafts are draft documents valid for a maximum of six
and may be updated, replaced, or obsoleted by other documents at any months and may be updated, replaced, or obsoleted by other documents
time. It is inappropriate to use Internet- Drafts as reference at any time. It is inappropriate to use Internet- Drafts as
material or to cite them other than as "work in progress." reference material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html. http://www.ietf.org/shadow.html.
Abstract Abstract
This document proposes HighSpeed TCP, a modification to TCP's This document proposes HighSpeed TCP, a modification to TCP's
congestion control mechanism for use with TCP connections with large congestion control mechanism for use with TCP connections with
congestion windows. The congestion control mechanisms of the current large congestion windows. The congestion control mechanisms
Standard TCP constrains the congestion windows that can be achieved of the current Standard TCP constrains the congestion windows
by TCP in realistic environments. For example, for a Standard TCP that can be achieved by TCP in realistic environments. For
connection with 1500-byte packets and a 100 ms round-trip time, example, for a Standard TCP connection with 1500-byte packets
achieving a steady-state throughput of 10 Gbps would require an and a 100 ms round-trip time, achieving a steady-state
average congestion window of 83,333 segments, and a packet drop rate throughput of 10 Gbps would require an average congestion
of at most one congestion event every 5,000,000,000 packets (or window of 83,333 segments, and a packet drop rate of at most
equivalently, at most one congestion event every 1 2/3 hours). This one congestion event every 5,000,000,000 packets (or
is widely acknowledged as an unrealistic constraint. To address this equivalently, at most one congestion event every 1 2/3 hours).
limitation of TCP, this document proposes HighSpeed TCP, and solicits This is widely acknowledged as an unrealistic constraint. To
experimentation and feedback from the wider community. address this limitation of TCP, this document proposes
HighSpeed TCP, and solicits experimentation and feedback from
the wider community.
TO BE DELETED BY THE RFC EDITOR UPON PUBLICATION: TO BE DELETED BY THE RFC EDITOR UPON PUBLICATION:
Changes from draft-floyd-tcp-highspeed-01.txt: Changes from draft-floyd-tcp-highspeed-02.txt:
* Added a section on "Tradeoffs for Choosing Congestion Control * Added a section on "Deployment issues."
Parameters".
* Added mention of Scalable TCP from Tom Kelly. * Added a short section on "Implementation issues."
Changes from draft-floyd-tcp-highspeed-00.txt: * Added a section on "Limiting burstiness on short time
scales".
* Added a discussion on related work about changing the PMTU. * Added to the discussion on convergence times.
* Added a discussion of an alternate, linear response function. * Clarified that "log" is "log base 10".
* Added a discussion of the TCP window scale option. * Clarified that W = Low_window and W_1 = High_window, in the
equation for b(w).
* Added a discussion of HighSpeed TCP as roughly emulating the Changes from draft-floyd-tcp-highspeed-01.txt:
congestion control response of N parallel TCP connections.
* Added a discussion of the time to converge to fairness. * Added a section on "Tradeoffs for Choosing Congestion
Control Parameters".
* Expanded the Introduction. * Added mention of Scalable TCP from Tom Kelly.
Changes from draft-floyd-tcp-highspeed-00.txt:
* Added a discussion on related work about changing the PMTU.
* Added a discussion of an alternate, linear response
function.
* Added a discussion of the TCP window scale option.
* Added a discussion of HighSpeed TCP as roughly emulating the
congestion control response of N parallel TCP connections.
* Added a discussion of the time to converge to fairness.
* Expanded the Introduction.
Table of Contents
1. Introduction. . . . . . . . . . . . . . . . . . . . . . 5
2. The Problem Description.. . . . . . . . . . . . . . . . 6
3. Design Guidelines.. . . . . . . . . . . . . . . . . . . 6
4. Non-Goals.. . . . . . . . . . . . . . . . . . . . . . . 7
5. Modifying the TCP Response Function.. . . . . . . . . . 8
6. Fairness Implications of the HighSpeed Response
Function.. . . . . . . . . . . . . . . . . . . . . . . . . 11
7. Translating the HighSpeed Response Function into
Congestion Control Parameters. . . . . . . . . . . . . . . 14
8. An alternate, linear response functions.. . . . . . . . 16
9. Tradeoffs for Choosing Congestion Control Parame-
ters.. . . . . . . . . . . . . . . . . . . . . . . . . . . 19
9.1. The Number of Round-Trip Times between Loss
Events. . . . . . . . . . . . . . . . . . . . . . . . . . 19
9.2. The Number of Packet Drops per Loss Event,
with Drop-Tail. . . . . . . . . . . . . . . . . . . . . . 19
10. Related Issues . . . . . . . . . . . . . . . . . . . . 20
10.1. Slow-Start. . . . . . . . . . . . . . . . . . . . . 20
10.2. Limiting burstiness on short time scales. . . . . . 21
10.3. Other limitations on window size. . . . . . . . . . 22
10.4. Implementation issues.. . . . . . . . . . . . . . . 22
11. Deployment issues. . . . . . . . . . . . . . . . . . . 22
11.1. Deployment issues of HighSpeed TCP. . . . . . . . . 22
11.2. Deployment issues of Scalable TCP . . . . . . . . . 24
12. Related Work in HighSpeed TCP. . . . . . . . . . . . . 26
13. Relationship to other Work.. . . . . . . . . . . . . . 27
14. Conclusions. . . . . . . . . . . . . . . . . . . . . . 28
15. Acknowledgements . . . . . . . . . . . . . . . . . . . 28
16. Normative References . . . . . . . . . . . . . . . . . 28
17. Informative References . . . . . . . . . . . . . . . . 28
18. Security Considerations. . . . . . . . . . . . . . . . 31
19. IANA Considerations. . . . . . . . . . . . . . . . . . 31
20. TCP's Loss Event Rate in Steady-State. . . . . . . . . 31
1. Introduction. 1. Introduction.
This document proposes HighSpeed TCP, a modification to TCP's This document proposes HighSpeed TCP, a modification to TCP's
congestion control mechanism for use with TCP connections with large congestion control mechanism for use with TCP connections with large
congestion windows. In a steady-state environment, with a packet congestion windows. In a steady-state environment, with a packet
loss rate p, the current Standard TCP's average congestion window is loss rate p, the current Standard TCP's average congestion window is
roughly 1.2/sqrt(p) segments. This places a serious constraint on roughly 1.2/sqrt(p) segments. This places a serious constraint on
the congestion windows that can be achieved by TCP in realistic the congestion windows that can be achieved by TCP in realistic
environments. For example, for a Standard TCP connection with environments. For example, for a Standard TCP connection with
1500-byte packets and a 100 ms round-trip time, achieving a steady- 1500-byte packets and a 100 ms round-trip time, achieving a steady-
state throughput of 10 Gbps would require an average congestion state throughput of 10 Gbps would require an average congestion
window of 83,333 segments, and a packet drop rate of at most one window of 83,333 segments, and a packet drop rate of at most one
congestion event every 5,000,000,000 packets (or equivalently, at congestion event every 5,000,000,000 packets (or equivalently, at
most one congestion event every 1 2/3 hours). most one congestion event every 1 2/3 hours). The average packet
drop rate of at most 2*10^(-10) needed for full link utilization in
this environment corresponds to a bit error rate of at most
2*10^(-14), and this is an unrealistic requirement for current
networks.
To address this fundamental limitation of TCP and of the TCP response To address this fundamental limitation of TCP and of the TCP
function (the function mapping the steady-state packet drop rate to response function (the function mapping the steady-state packet drop
TCP's average sending rate in packets per round-trip time), this rate to TCP's average sending rate in packets per round-trip time),
document proposes modifying the TCP response function for regimes this document describes a modified TCP response function for regimes
with higher congestion windows. This document also solicits with higher congestion windows. This document also solicits
experimentation and feedback on HighSpeed TCP from the wider experimentation and feedback on HighSpeed TCP from the wider
community. community.
Because HighSpeed TCP's modified response function would only take Because HighSpeed TCP's modified response function would only take
effect with higher congestion windows, HighSpeed TCP does not modify effect with higher congestion windows, HighSpeed TCP does not modify
TCP behavior in environments with mild to heavy congestion, and TCP behavior in environments with mild to heavy congestion, and
therefore does not introduce any new dangers of congestion collapse. therefore does not introduce any new dangers of congestion collapse.
However, if relative fairness between HighSpeed TCP connections is to However, if relative fairness between HighSpeed TCP connections is
be preserved, then in our view any modification to the TCP response to be preserved, then in our view any modification to the TCP
function should be globally-agreed-upon in the IETF, rather than made response function should be addressed in the IETF, rather than made
as ad hoc decisions by individual implementors or TCP senders. as ad hoc decisions by individual implementors or TCP senders.
Modifications to the TCP response function would also have Modifications to the TCP response function would also have
implications for transport protocols that use TFRC and other forms of implications for transport protocols that use TFRC and other forms
equation-based congestion control, as these congestion control of equation-based congestion control, as these congestion control
mechanisms directly use the TCP response function [TFRC]. mechanisms directly use the TCP response function [RFC3448].
This proposal for HighSpeed TCP focuses specifically on a proposed This proposal for HighSpeed TCP focuses specifically on a proposed
change to the TCP response function, and its implications for TCP. change to the TCP response function, and its implications for TCP.
This document does not address what we view as a separate fundamental This document does not address what we view as a separate
issue, of the mechanisms required to enable best-effort connections fundamental issue, of the mechanisms required to enable best-effort
to *start* with large initial windows. In our view, while HighSpeed connections to *start* with large initial windows. In our view,
TCP proposes a somewhat fundamental change to the TCP response while HighSpeed TCP proposes a somewhat fundamental change to the
function, at the same time it is a relatively simple change to TCP response function, at the same time it is a relatively simple
implement in a single TCP sender, and presents no dangers in terms of change to implement in a single TCP sender, and presents no dangers
congestion collapse. In contrast, in our view, the problem of in terms of congestion collapse. In contrast, in our view, the
enabling connections to *start* with large initial windows is problem of enabling connections to *start* with large initial
inherently more risky and structurally more difficult, requiring some windows is inherently more risky and structurally more difficult,
form of explicit feedback from all of the routers along the path. requiring some form of explicit feedback from all of the routers
This is another reason why we would propose addressing the problem of along the path. This is another reason why we would propose
starting with large initial windows separately, and on a separate addressing the problem of starting with large initial windows
timetable, from the problem of modifying the TCP response function. separately, and on a separate timetable, from the problem of
modifying the TCP response function.
2. The Problem Description. 2. The Problem Description.
This section describes the number of round-trip times between This section describes the number of round-trip times between
congestion events required for a Standard TCP flow to achieve an congestion events required for a Standard TCP flow to achieve an
average throughput of B bps, given packets of D bytes and a round- average throughput of B bps, given packets of D bytes and a round-
trip time of R seconds. A congestion event refers to a window of trip time of R seconds. A congestion event refers to a window of
data with one or more dropped or ECN-marked packets (where ECN stands data with one or more dropped or ECN-marked packets (where ECN
for Explicit Congestion Notification). stands for Explicit Congestion Notification).
From Appendix A, achieving an average TCP throughput of B bps From Appendix A, achieving an average TCP throughput of B bps
requires a loss event at most every BR/(12D) round-trip times. This requires a loss event at most every BR/(12D) round-trip times. This
is illustrated in Table 1, for R = 0.1 seconds and D = 1500 bytes. is illustrated in Table 1, for R = 0.1 seconds and D = 1500 bytes.
The table also gives the average congestion window W of BR/(8D), and The table also gives the average congestion window W of BR/(8D), and
the steady-state packet drop rate P of 1.5/W^2. the steady-state packet drop rate P of 1.5/W^2.
TCP Throughput (Mbps) RTTs Between Losses W P TCP Throughput (Mbps) RTTs Between Losses W P
--------------------- ------------------- ------ ----- --------------------- ------------------- ---- -----
1 5.5 8.3 0.02 1 5.5 8.3 0.02
10 55.5 83.3 0.0002 10 55.5 83.3 0.0002
100 555.5 833.3 0.000002 100 555.5 833.3 0.000002
1000 5555.5 8333.3 0.00000002 1000 5555.5 8333.3 0.00000002
10000 55555.5 83333.3 0.0000000002 10000 55555.5 83333.3 0.0000000002
Table 1: RTTs Between Congestion Events for Standard TCP, for Table 1: RTTs Between Congestion Events for Standard TCP, for
1500-Byte Packets and a Round-Trip Time of 0.1 Seconds. 1500-Byte Packets and a Round-Trip Time of 0.1 Seconds.
This document proposes HighSpeed TCP, a minimal modification to TCP's This document proposes HighSpeed TCP, a minimal modification to
increase and decrease parameters, for TCP connections with larger TCP's increase and decrease parameters, for TCP connections with
congestion windows, to allow TCP to achieve high throughput with more larger congestion windows, to allow TCP to achieve high throughput
realistic requirements for the steady-state packet drop rate. with more realistic requirements for the steady-state packet drop
Equivalently, HighSpeed TCP has more realistic requirements for the rate. Equivalently, HighSpeed TCP has more realistic requirements
number of round-trip times between loss events. for the number of round-trip times between loss events.
3. Design Guidelines. 3. Design Guidelines.
Our proposal for HighSpeed TCP is motivated by the following Our proposal for HighSpeed TCP is motivated by the following
requirements: requirements:
* Achieve high per-connection throughput without requiring * Achieve high per-connection throughput without requiring
unrealistically low packet loss rates. unrealistically low packet loss rates.
* Reach high throughput reasonably quickly when in slow-start. * Reach high throughput reasonably quickly when in slow-start.
* Reach high throughput without overly long delays when recovering * Reach high throughput without overly long delays when recovering
from multiple retransmit timeouts, or when ramping-up from a period from multiple retransmit timeouts, or when ramping-up from a period
with small congestion windows. with small congestion windows.
* No additional feedback or support required from routers: * No additional feedback or support required from routers:
For example, the goal is for acceptable performance in both ECN- For example, the goal is for acceptable performance in both ECN-
capable and non-ECN-capable environments, and with Drop-Tail as well capable and non-ECN-capable environments, and with Drop-Tail as well
as with Active Queue Management such as RED in the routers. as with Active Queue Management such as RED in the routers.
* No additional feedback required from TCP receivers. * No additional feedback required from TCP receivers.
* TCP-compatible performance in environments with moderate or high * TCP-compatible performance in environments with moderate or high
congestion: congestion:
Equivalently, the requirement is that there be no additional load on Equivalently, the requirement is that there be no additional load on
the network (in terms of increased packet drop rates) in environments the network (in terms of increased packet drop rates) in
with moderate or high congestion. environments with moderate or high congestion.
* Performance at least as good as Standard TCP in environments with * Performance at least as good as Standard TCP in environments with
moderate or high congestion. moderate or high congestion.
* Acceptable transient performance, in terms of increases in the * Acceptable transient performance, in terms of increases in the
congestion window in one round-trip time, responses to severe congestion window in one round-trip time, responses to severe
congestion, and convergence times to fairness. congestion, and convergence times to fairness.
Currently, users wishing to achieve throughputs of 1Gbps or more Currently, users wishing to achieve throughputs of 1 Gbps or more
typically open up multiple TCP connections in parallel, or use MulTCP typically open up multiple TCP connections in parallel, or use
[CO98,GRK99], which behaves roughly like the aggregate of N virtual MulTCP [CO98,GRK99], which behaves roughly like the aggregate of N
TCP connections. While this approach suffices for the occasional virtual TCP connections. While this approach suffices for the
user on well-provisioned links, it leaves the parameter N to be occasional user on well-provisioned links, it leaves the parameter N
determined by the user, and results in more aggressive performance to be determined by the user, and results in more aggressive
and higher steady-state packet drop rates if used in environments performance and higher steady-state packet drop rates if used in
with periods of moderate or high congestion. We believe that a new environments with periods of moderate or high congestion. We
approach is needed that offers more flexibility, more effectively believe that a new approach is needed that offers more flexibility,
scales to a wide range of available bandwidths, and competes more more effectively scales to a wide range of available bandwidths, and
fairly with Standard TCP in congested environments. competes more fairly with Standard TCP in congested environments.
4. Non-Goals. 4. Non-Goals.
The following are explicitly *not* goals of our work: The following are explicitly *not* goals of our work:
* Non-goal: TCP-compatible performance in environments with very low * Non-goal: TCP-compatible performance in environments with very low
packet drop rates. packet drop rates.
We note that our proposal does not require, or deliver, TCP- We note that our proposal does not require, or deliver, TCP-
compatible performance in environments with very low packet drop compatible performance in environments with very low packet drop
rates, e.g., with packet loss rates of 10^-5 or 10^-6. As we discuss rates, e.g., with packet loss rates of 10^-5 or 10^-6. As we
later in this document, we assume that Standard TCP is unable to make discuss later in this document, we assume that Standard TCP is
effective use of the available bandwidth in environments with loss unable to make effective use of the available bandwidth in
rates of 10^-6 in any case, so that it is acceptable and appropriate environments with loss rates of 10^-6 in any case, so that it is
for HighSpeed TCP to perform more aggressively than Standard TCP is acceptable and appropriate for HighSpeed TCP to perform more
such an environment. aggressively than Standard TCP is such an environment.
* Non-goal: Ramping-up more quickly than allowed by slow-start. * Non-goal: Ramping-up more quickly than allowed by slow-start.
It is our belief that ramping-up more quickly than allowed by slow- It is our belief that ramping-up more quickly than allowed by slow-
start would necessitate more explicit feedback from routers along the start would necessitate more explicit feedback from routers along
path. The proposal for HighSpeed TCP is focused on changes to TCP the path. The proposal for HighSpeed TCP is focused on changes to
that could be effectively deployed in the current Internet TCP that could be effectively deployed in the current Internet
environment. environment.
* Non-goal: Avoiding oscillations in environments with only one-way, * Non-goal: Avoiding oscillations in environments with only one-way,
long-lived flows all with the same round-trip times. long-lived flows all with the same round-trip times.
While we agree that attention to oscillatory behavior is useful, While we agree that attention to oscillatory behavior is useful,
avoiding oscillations in aggregate throughput has not been our avoiding oscillations in aggregate throughput has not been our
primary consideration, particularly for simplified environments primary consideration, particularly for simplified environments
limited to one-way, long-lived flows all with the same, large round- limited to one-way, long-lived flows all with the same, large round-
trip times. Our assessment is that some oscillatory behavior in trip times. Our assessment is that some oscillatory behavior in
these extreme environments is an acceptable price to pay for the these extreme environments is an acceptable price to pay for the
other benefits of HighSpeed TCP. other benefits of HighSpeed TCP.
5. Modifying the TCP Response Function. 5. Modifying the TCP Response Function.
The TCP response function, w = 1.2/sqrt(p), gives TCP's average The TCP response function, w = 1.2/sqrt(p), gives TCP's average
congestion window w in MSS-sized segments, as a function of the congestion window w in MSS-sized segments, as a function of the
steady-state packet drop rate p [FF98]. This TCP response function steady-state packet drop rate p [FF98]. This TCP response function
is a direct consequence of TCP's Additive Increase Multiplicative is a direct consequence of TCP's Additive Increase Multiplicative
Decrease (AIMD) mechanisms of increasing the congestion window by Decrease (AIMD) mechanisms of increasing the congestion window by
roughly one segment per round-trip time in the absence of congestion, roughly one segment per round-trip time in the absence of
and halving the congestion window in response to a round-trip time congestion, and halving the congestion window in response to a
with a congestion event. This response function for Standard TCP is round-trip time with a congestion event. This response function for
reflected in the table below. In this proposal we restrict our Standard TCP is reflected in the table below. In this proposal we
attention to TCP performance in environments with packet loss rates restrict our attention to TCP performance in environments with
of at most 10^-2, and so we can ignore the more complex response packet loss rates of at most 10^-2, and so we can ignore the more
functions that are required to model TCP performance in more complex response functions that are required to model TCP
congested environments with retransmit timeouts. From Appendix A, an performance in more congested environments with retransmit timeouts.
average congestion window of W corresponds to an average of W/1.5 From Appendix A, an average congestion window of W corresponds to an
round-trip times between loss events for Standard TCP. average of 2/3 W round-trip times between loss events for Standard
TCP (with the congestion window varying from 2/3 W to 4/3 W).
Packet Drop Rate P Congestion Window W RTTs Between Losses Packet Drop Rate P Congestion Window W RTTs Between Losses
------------------ ------------------- ------------------- ------------------ ------------------- -------------------
10^-2 12 8 10^-2 12 8
10^-3 38 25 10^-3 38 25
10^-4 120 80 10^-4 120 80
10^-5 379 252 10^-5 379 252
10^-6 1200 800 10^-6 1200 800
10^-7 3795 2530 10^-7 3795 2530
10^-8 12000 8000 10^-8 12000 8000
10^-9 37948 25298 10^-9 37948 25298
10^-10 120000 80000 10^-10 120000 80000
Table 2: TCP Response Function for Standard TCP. The average Table 2: TCP Response Function for Standard TCP. The average
congestion window W in MSS-sized segments is given as a function of congestion window W in MSS-sized segments is given as a function of
the packet drop rate P. the packet drop rate P.
To specify a modified response function for HighSpeed TCP, we use To specify a modified response function for HighSpeed TCP, we use
three parameters, Low_Window, High_Window, and High_P. To ensure TCP three parameters, Low_Window, High_Window, and High_P. To ensure
compatibility, the HighSpeed response function uses the same response TCP compatibility, the HighSpeed response function uses the same
function as Standard TCP when the current congestion window is at response function as Standard TCP when the current congestion window
most Low_Window, and uses the HighSpeed response function when the is at most Low_Window, and uses the HighSpeed response function when
current congestion window is greater than Low_Window. In this the current congestion window is greater than Low_Window. In this
document we set Low_Window to 38 MSS-sized segments, corresponding to document we set Low_Window to 38 MSS-sized segments, corresponding
a packet drop rate of 10^-3 for TCP. to a packet drop rate of 10^-3 for TCP.
To specify the upper end of the HighSpeed response function, we To specify the upper end of the HighSpeed response function, we
specify the packet drop rate needed in the HighSpeed response specify the packet drop rate needed in the HighSpeed response
function to achieve an average congestion window of 83000 segments. function to achieve an average congestion window of 83000 segments.
This is roughly the window needed to sustain 10Gbps throughput, for a This is roughly the window needed to sustain 10 Gbps throughput, for
TCP connection with the default packet size and round-trip time used a TCP connection with the default packet size and round-trip time
earlier in this document. For High_Window set to 83000, we specify used earlier in this document. For High_Window set to 83000, we
High_P of 10^-7; that is, with HighSpeed TCP a packet drop rate of specify High_P of 10^-7; that is, with HighSpeed TCP a packet drop
10^-7 allows the HighSpeed TCP connection to achieve an average rate of 10^-7 allows the HighSpeed TCP connection to achieve an
congestion window of 83000 segments. We believe that this loss rate average congestion window of 83000 segments. We believe that this
sets an achieveable target for high-speed environments, while still loss rate sets an achievable target for high-speed environments,
allowing acceptable fairness for the HighSpeed response function when while still allowing acceptable fairness for the HighSpeed response
competing with Standard TCP in environments with packet drop rates of function when competing with Standard TCP in environments with
10^-4 or 10^5. packet drop rates of 10^-4 or 10^5.
For simplicity, for the HighSpeed response function we maintain the For simplicity, for the HighSpeed response function we maintain the
property that the response function gives a straight line on a log- property that the response function gives a straight line on a log-
log scale (as does the response function for Standard TCP, for low to log scale (as does the response function for Standard TCP, for low
moderate congestion). This results in the following response to moderate congestion). This results in the following response
function, for values of the average congestion window W greater than function, for values of the average congestion window W greater than
Low_Window: Low_Window:
W = (p/Low_P)^S Low_Window, W = (p/Low_P)^S Low_Window,
for Low_P the packet drop rate corresponding to Low_Window, and for S for Low_P the packet drop rate corresponding to Low_Window, and for
as following constant [FRS02]: S as following constant [FRS02]:
S = (log High_Window - log Low_Window)/(log High_P - log Low_P). S = (log High_Window - log Low_Window)/(log High_P - log Low_P).
For example, for Low_Window set to 38, we have Low_P of 10^-3 (for (In this paper, "log x" refers to the log base 10.) For example,
compatibility with Standard TCP). Thus, for High_Window set to 83000 for Low_Window set to 38, we have Low_P of 10^-3 (for compatibility
and High_P set to 10^-7, we get the following response function: with Standard TCP). Thus, for High_Window set to 83000 and High_P
set to 10^-7, we get the following response function:
W = 0.12/p^0.835. (1) W = 0.12/p^0.835. (1)
This HighSpeed response function is illustrated in Table 3 below. This HighSpeed response function is illustrated in Table 3 below.
For HighSpeed TCP, the number of round-trip times between losses, For HighSpeed TCP, the number of round-trip times between losses,
1/(pW), equals 12.7 W^0.2, for W > 38 segments. 1/(pW), equals 12.7 W^0.2, for W > 38 segments.
Packet Drop Rate P Congestion Window W RTTs Between Losses Packet Drop Rate P Congestion Window W RTTs Between Losses
------------------ ------------------- ------------------- ------------------ ------------------- -------------------
10^-2 12 8 10^-2 12 8
10^-3 38 25 10^-3 38 25
10^-4 263 38 10^-4 263 38
10^-5 1795 57 10^-5 1795 57
10^-6 12279 83 10^-6 12279 83
10^-7 83981 123 10^-7 83981 123
10^-8 574356 180 10^-8 574356 180
10^-9 3928088 264 10^-9 3928088 264
10^-10 26864653 388 10^-10 26864653 388
Table 3: TCP Response Function for HighSpeed TCP. The average Table 3: TCP Response Function for HighSpeed TCP. The average
congestion window W in MSS-sized segments is given as a function of congestion window W in MSS-sized segments is given as a function of
the packet drop rate P. the packet drop rate P.
We believe that the problem of backward compatibility with Standard We believe that the problem of backward compatibility with Standard
TCP requires a response function that is quite close to that of TCP requires a response function that is quite close to that of
Standard TCP for loss rates of 10^-1, 10^-2, or 10^-3. We believe, Standard TCP for loss rates of 10^-1, 10^-2, or 10^-3. We believe,
however, that such stringent TCP-compatibility is not required for however, that such stringent TCP-compatibility is not required for
smaller loss rates, and that an appropriate response function is one smaller loss rates, and that an appropriate response function is one
that gives a plausible packet drop rate for a connection throughput that gives a plausible packet drop rate for a connection throughput
of 10Gbps. This also gives a slowly increasing number of round-trip of 10 Gbps. This also gives a slowly increasing number of round-
times between loss events as a function of a decreasing packet drop trip times between loss events as a function of a decreasing packet
rate. drop rate.
Another way to look at the HighSpeed response function is to consider Another way to look at the HighSpeed response function is to
that HighSpeed TCP is roughly emulating the congestion control consider that HighSpeed TCP is roughly emulating the congestion
response of N parallel TCP connections, where N is initially one, and control response of N parallel TCP connections, where N is initially
where N increases as a function of the HighSpeed TCP's congestion one, and where N increases as a function of the HighSpeed TCP's
window. Thus for the HighSpeed response function in Equation (1) congestion window. Thus for the HighSpeed response function in
above, the response function can be viewed as equivalent to that of Equation (1) above, the response function can be viewed as
N(W) parallel TCP connections, where N(W) varies as a function of the equivalent to that of N(W) parallel TCP connections, where N(W)
congestion window W. Recall that for a single standard TCP varies as a function of the congestion window W. Recall that for a
connection, the average congestion window equals 1.2/sqrt(p). For N single standard TCP connection, the average congestion window equals
parallel TCP connections, the aggregate congestion window W_n equals 1.2/sqrt(p). For N parallel TCP connections, the aggregate
N*1.2/sqrt(p). From the HighSpeed response function in Equation (1) congestion window for the N connections equals N*1.2/sqrt(p). From
and the relationship above, we can derive the following: the HighSpeed response function in Equation (1) and the relationship
above, we can derive the following:
N(W) = 0.23*W^(0.4) N(W) = 0.23*W^(0.4)
for N(W) the number of parallel TCP connections emulated by the for N(W) the number of parallel TCP connections emulated by the
HighSpeed TCP response function, and for N(W) >= 1. This is shown in HighSpeed TCP response function, and for N(W) >= 1. This is shown
Table 4 below. in Table 4 below.
Congestion Window W Number N(W) of Parallel TCPs Congestion Window W Number N(W) of Parallel TCPs
------------------- ------------------------- ------------------- -------------------------
1 1 1 1
10 1 10 1
100 1.4 100 1.4
1,000 3.6 1,000 3.6
10,000 9.2 10,000 9.2
100,000 23.0 100,000 23.0
Table 4: Number N(W) of parallel TCP connections roughly emulated by Table 4: Number N(W) of parallel TCP connections roughly emulated by
the HighSpeed TCP response function. the HighSpeed TCP response function.
We do not in this document attempt to seriously evaluate the We do not in this document attempt to seriously evaluate the
HighSpeed response function for congestion windows greater than HighSpeed response function for congestion windows greater than
100,000 packets. We believe that we will learn more about the 100,000 packets. We believe that we will learn more about the
requirements for sustaining the throughput of best-effort connections requirements for sustaining the throughput of best-effort
in that range as we gain more experience with HighSpeed TCP with connections in that range as we gain more experience with HighSpeed
congestion windows of thousands and tens of thousands of packets. TCP with congestion windows of thousands and tens of thousands of
There also might be limitations to the per-connection throughput that packets. There also might be limitations to the per-connection
can be realistically achieved for best-effort traffic in the absence throughput that can be realistically achieved for best-effort
of additional support or feedback from the routers along the path. traffic, in terms of congestion window of hundreds of thousands of
packets or more, in the absence of additional support or feedback
from the routers along the path.
6. Fairness Implications of the HighSpeed Response Function. 6. Fairness Implications of the HighSpeed Response Function.
The Standard and Highspeed Response Functions can be used directly to The Standard and Highspeed Response Functions can be used directly
infer the relative fairness between flows using the two response to infer the relative fairness between flows using the two response
functions. For example, given a packet drop rate P, assume that functions. For example, given a packet drop rate P, assume that
Standard TCP has an average congestion window of W_Standard, and Standard TCP has an average congestion window of W_Standard, and
HighSpeed TCP has a higher average congestion window of W_HighSpeed. HighSpeed TCP has a higher average congestion window of W_HighSpeed.
In this case, a single HighSpeed TCP connection is receiving
W_HighSpeed/W_Standard times the throughput of a single Standard TCP
connection competing in the same environment.
This relative fairness is illustrated below in Table 5, for the In this case, a single HighSpeed TCP connection is receiving
parameters used for the Highspeed response function in the section W_HighSpeed/W_Standard times the throughput of a single Standard TCP
above. The second column gives the relative fairness, for the connection competing in the same environment.
steady-state packet drop rate specified in the first column. To help
calibrate, the third column gives the aggregate average congestion
window for the two TCP connections, and the fourth column gives the
bandwidth that would be needed by the two connections to achieve that
aggregate window and packet drop rate, given 100 ms round-trip times
and 1500-byte packets.
Packet Drop Rate P Fairness Aggregate Window Bandwidth This relative fairness is illustrated below in Table 5, for the
------------------ -------- ---------------- --------- parameters used for the Highspeed response function in the section
10^-2 1.0 24 2.8 Mbps above. The second column gives the relative fairness, for the
10^-3 1.0 76 9.1 Mbps steady-state packet drop rate specified in the first column. To
10^-4 2.2 383 45.9 Mbps help calibrate, the third column gives the aggregate average
10^-5 4.7 2174 260.8 Mbps congestion window for the two TCP connections, and the fourth column
10^-6 10.2 13479 1.6 Gbps gives the bandwidth that would be needed by the two connections to
10^-7 22.1 87776 10.5 Gbps achieve that aggregate window and packet drop rate, given 100 ms
10^-8 47.9 586356 70.3 Gbps round-trip times and 1500-byte packets.
10^-9 103.5 3966036 475.9 Gbps
10^-10 223.9 26984653 3238.1 Gbps
Table 5: Relative Fairness between the HighSpeed and Standard Packet Drop Rate P Fairness Aggregate Window Bandwidth
Response Functions. ------------------ -------- ---------------- ---------
10^-2 1.0 24 2.8 Mbps
10^-3 1.0 76 9.1 Mbps
10^-4 2.2 383 45.9 Mbps
10^-5 4.7 2174 260.8 Mbps
10^-6 10.2 13479 1.6 Gbps
10^-7 22.1 87776 10.5 Gbps
Thus, for packet drop rates of 10^-4, a flow with the HighSpeed Table 5: Relative Fairness between the HighSpeed and Standard
response function can expect to receive 2.2 times the throughput of a Response Functions.
flow using the Standard response function, given the same round-trip
times and packet sizes. With packet drop rates of 10^-6 (or 10^-7),
the unfairness is more severe, and we have entered the regime where a
Standard TCP connection requires a congestion event at most every 800
(or 2530) round-trip times in order to make use of the available
bandwidth. Our judgement would be that there are not a lot of TCP
connections effectively operating in this regime today, with
congestion windows of thousands of packets, and that therefore the
benefits of the HighSpeed response function would outweigh the
unfairness that would be experienced by Standard TCP in this regime.
However, one purpose of this document is to solicit feedback on this
issue. The parameter Low_Window determines directly the point of
divergence between the Standard and HighSpeed Response Functions.
The third column of Table 5, the Aggregate Window, gives the Thus, for packet drop rates of 10^-4, a flow with the HighSpeed
aggregate congestion window of the two competing TCP connections, response function can expect to receive 2.2 times the throughput of
with HighSpeed and Standard TCP, given the packet drop rate specified a flow using the Standard response function, given the same round-
in the first column. From Table 5, a HighSpeed TCP connection would trip times and packet sizes. With packet drop rates of 10^-6 (or
receive ten times the bandwidth of a Standard TCP in an environment 10^-7), the unfairness is more severe, and we have entered the
with a packet drop rate of 10^-6. This would occur when the two regime where a Standard TCP connection requires at most one
flows sharing a single pipe achieved an aggregate window of 13479 congestion event every 800 (or 2530) round-trip times in order to
packets. Given a round-trip time of 100 ms and a packet size of 1500 make use of the available bandwidth. Our judgement would be that
bytes, this would occur with an available bandwidth for the two there are not a lot of TCP connections effectively operating in this
competing flows of 1.6 Gbps. regime today, with congestion windows of thousands of packets, and
that therefore the benefits of the HighSpeed response function would
outweigh the unfairness that would be experienced by Standard TCP in
this regime. However, one purpose of this document is to solicit
feedback on this issue. The parameter Low_Window determines
directly the point of divergence between the Standard and HighSpeed
Response Functions.
Next we consider the time that it takes two HighSpeed TCP flows to The third column of Table 5, the Aggregate Window, gives the
converge to fairness. The worst case for convergence to fairness aggregate congestion window of the two competing TCP connections,
occurs when a new flow is starting up, competing against a high- with HighSpeed and Standard TCP, given the packet drop rate
bandwidth existing flow, and the new flow suffers a packet drop and specified in the first column. From Table 5, a HighSpeed TCP
exits slow-start while its window is still small. In the worst case, connection would receive ten times the bandwidth of a Standard TCP
consider that the new flow has entered the congestion avoidance phase in an environment with a packet drop rate of 10^-6. This would
while its window is only one packet. A standard TCP flow in occur when the two flows sharing a single pipe achieved an aggregate
congestion avoidance increases its window by at most one packet per window of 13479 packets. Given a round-trip time of 100 ms and a
round-trip time, and after N round-trip times has only achieved a packet size of 1500 bytes, this would occur with an available
window of N packets (when starting with a window of 1 in the first bandwidth for the two competing flows of 1.6 Gbps.
round-trip time). In contrast, a HighSpeed TCP flows increases much
faster than a standard TCP flow while in the congestion avoidance
phase, and we can expect its convergence to fairness to be much
better. This is shown in Table 6 below. The script used to generate
this table is given in Appendix C.
RTT HS_Window Standard_TCP_Window Next we consider the time that it takes a standard or HighSpeed TCP
--- --------- ------------------- flow to converge to fairness against a pre-existing HighSpeed TCP
100 131 100 flow. The worst case for convergence to fairness occurs when a new
200 475 200 flow is starting up, competing against a high-bandwidth existing
300 1131 300 flow, and the new flow suffers a packet drop and exits slow-start
400 2160 400 while its window is still small. In the worst case, consider that
500 3601 500 the new flow has entered the congestion avoidance phase while its
600 5477 600 window is only one packet. A standard TCP flow in congestion
700 7799 700 avoidance increases its window by at most one packet per round-trip
800 10567 800 time, and after N round-trip times has only achieved a window of N
900 13774 900 packets (when starting with a window of 1 in the first round-trip
1000 17409 1000 time). In contrast, a HighSpeed TCP flows increases much faster
1100 21455 1100 than a standard TCP flow while in the congestion avoidance phase,
1200 25893 1200 and we can expect its convergence to fairness to be much better.
1300 30701 1300 This is shown in Table 6 below. The script used to generate this
1400 35856 1400 table is given in Appendix C.
1500 41336 1500
1600 47115 1600
1700 53170 1700
1800 59477 1800
1900 66013 1900
2000 72754 2000
Table 6: For a HighSpeed and a Standard TCP connection, the RTT HS_Window Standard_TCP_Window
congestion window during congestion avoidance phase (starting with a --- --------- -------------------
congestion window of 1 packet during RTT 1. 100 131 100
200 475 200
300 1131 300
400 2160 400
500 3601 500
600 5477 600
700 7799 700
800 10567 800
900 13774 900
1000 17409 1000
1100 21455 1100
1200 25893 1200
1300 30701 1300
1400 35856 1400
1500 41336 1500
1600 47115 1600
1700 53170 1700
1800 59477 1800
1900 66013 1900
2000 72754 2000
The classic paper on relative fairness is from Chiu and Jain [CJ89]. Table 6: For a HighSpeed and a Standard TCP connection, the
This paper shows that AIMD (Additive Increase Multiplicative congestion window during congestion avoidance phase (starting with a
Decrease) converges to fairness in an environment with synchronized congestion window of 1 packet during RTT 1.
congestion events. From [CJ89], it is easy to see that MIMD and AIAD
do not converge to fairness in this environment. However, the
results of [CJ89] do not apply to an asynchronous environment such as
that of the current Internet, where the frequency of congestion
feedback can be different for different flows. For example, it has
been shown that MIMD converges to fair states in a model with
proportional instead of synchronous feedback in terms of packet drops
[GV02]. Thus, we are not concerned about abandoning a strict model The classic paper on relative fairness is from Chiu and Jain [CJ89].
of AIMD for HighSpeed TCP. This paper shows that AIMD (Additive Increase Multiplicative
Decrease) converges to fairness in an environment with synchronized
congestion events. From [CJ89], it is easy to see that MIMD and
AIAD do not converge to fairness in this environment. However, the
results of [CJ89] do not apply to an asynchronous environment such
as that of the current Internet, where the frequency of congestion
feedback can be different for different flows. For example, it has
been shown that MIMD converges to fair states in a model with
proportional instead of synchronous feedback in terms of packet
drops [GV02]. Thus, we are not concerned about abandoning a strict
model of AIMD for HighSpeed TCP.
7. Translating the HighSpeed Response Function into Congestion Control 7. Translating the HighSpeed Response Function into Congestion Control
Parameters. Parameters.
For equation-based congestion control such as TFRC, the HighSpeed For equation-based congestion control such as TFRC, the HighSpeed
Response Function above could be used directly by the TFRC congestion Response Function above could be used directly by the TFRC
control mechanism. However, for TCP the HighSpeed response function congestion control mechanism. However, for TCP the HighSpeed
would have to be translated into additive increase and multiplicative response function has to be translated into additive increase and
decrease parameters. The HighSpeed response function cannot be multiplicative decrease parameters. The HighSpeed response function
achieved by TCP with an additive increase of one segment per round- cannot be achieved by TCP with an additive increase of one segment
trip time and a multiplicative decrease of halving the current per round-trip time and a multiplicative decrease of halving the
congestion window; HighSpeed TCP will have to modify either the current congestion window; HighSpeed TCP will have to modify either
increase or the decrease parameter, or both. We have concluded that the increase or the decrease parameter, or both. We have concluded
HighSpeed TCP is most likely to achieve an acceptable compromise that HighSpeed TCP is most likely to achieve an acceptable
between moderate increases and timely decreases by modifying both the compromise between moderate increases and timely decreases by
increase and the decrease parameter. modifying both the increase and the decrease parameter.
That is, for HighSpeed TCP let the congestion window increase by a(w) That is, for HighSpeed TCP let the congestion window increase by
segments per round-trip time in the absence of congestion, and let a(w) segments per round-trip time in the absence of congestion, and
the congestion window decrease to w(1-b(w)) segments in response to a let the congestion window decrease to w(1-b(w)) segments in response
round-trip time with one or more loss events. Thus, in response to a to a round-trip time with one or more loss events. Thus, in
single acknowledgement HighSpeed TCP increases its congestion window response to a single acknowledgement HighSpeed TCP increases its
in segments as follows: congestion window in segments as follows:
w <- w + a(w)/w. w <- w + a(w)/w.
In response to a congestion event, HighSpeed TCP decreases as In response to a congestion event, HighSpeed TCP decreases as
follows: follows:
w <- (1-b(w))w. w <- (1-b(w))w.
For Standard TCP, a(w) = 1 and b(w) = 1/2, regardless of the value of For Standard TCP, a(w) = 1 and b(w) = 1/2, regardless of the value
w. HighSpeed TCP uses the same values of a(w) and b(w) for w <= of w. HighSpeed TCP uses the same values of a(w) and b(w) for w <=
Low_Window. This section specifies a(w) and b(w) for HighSpeed TCP Low_Window. This section specifies a(w) and b(w) for HighSpeed TCP
for larger values of w. for larger values of w.
For w = High_Window, we have specified a loss rate of High_P. From For w = High_Window, we have specified a loss rate of High_P. From
[FRS02], or from elementary calculations, this requires the following [FRS02], or from elementary calculations, this requires the
relationship between a(w) and b(w) for w = High_Window: following relationship between a(w) and b(w) for w = High_Window:
a(w) = High_Window^2 * High_P * 2 * b(w)/(2-b(w). (2) a(w) = High_Window^2 * High_P * 2 * b(w)/(2-b(w). (2)
We use the parameter High_Decrease to specify the decrease parameter We use the parameter High_Decrease to specify the decrease parameter
b(w) for w = High_Window, and use Equation (2) to derive the increase b(w) for w = High_Window, and use Equation (2) to derive the
parameter a(w) for w = High_Window. Along with High_P = 10^-7 and increase parameter a(w) for w = High_Window. Along with High_P =
High_Window = 83000, for example, we specify High_Decrease = 0.1, 10^-7 and High_Window = 83000, for example, we specify High_Decrease
specifying that b(83000) = 0.1, giving a decrease of 10% after a = 0.1, specifying that b(83000) = 0.1, giving a decrease of 10%
congestion event. Equation (2) then gives a(83000) = 72, for an after a congestion event. Equation (2) then gives a(83000) = 72,
increase of 72 segments, or just under 0.1%, within a round-trip for an increase of 72 segments, or just under 0.1%, within a round-
time, for w = 83000. trip time, for w = 83000.
This moderate decrease strikes us as acceptable, particularly when This moderate decrease strikes us as acceptable, particularly when
coupled with the role of TCP's ACK-clocking in limiting the sending coupled with the role of TCP's ACK-clocking in limiting the sending
rate in response to more severe congestion [BBFS01]. A more severe rate in response to more severe congestion [BBFS01]. A more severe
decrease would require a more aggressive increase in the congestion decrease would require a more aggressive increase in the congestion
window for a round-trip time without congestion. In particular, a window for a round-trip time without congestion. In particular, a
decrease factor High_Decrease of 0.5, as in Standard TCP, would decrease factor High_Decrease of 0.5, as in Standard TCP, would
require an increase of 459 segments per round-trip time when w = require an increase of 459 segments per round-trip time when w =
83000. 83000.
Given decrease parameters of b(w) = 1/2 for w = Low_Window, and b(w) Given decrease parameters of b(w) = 1/2 for w = Low_Window, and b(w)
= High_Decrease for w = High_Window, we are left to specify the value = High_Decrease for w = High_Window, we are left to specify the
of b(w) for other values of w > Low_Window. From [FRS02], we let value of b(w) for other values of w > Low_Window. From [FRS02], we
b(w) vary linearly as the log of w, as follows: let b(w) vary linearly as the log of w, as follows:
b(w) = (High_Decrease - 0.5) (log(w)-log(W)) / (log(W_1)-log(W)) + b(w) = (High_Decrease - 0.5) (log(w)-log(W)) / (log(W_1)-log(W)) +
0.5. 0.5,
The increase parameter a(w) can then be computed as follows: for W = Low_window and W_1 = High_window. The increase parameter
a(w) can then be computed as follows:
a(w) = w^2 * p(w) * 2 * b(w)/(2-b(w)), a(w) = w^2 * p(w) * 2 * b(w)/(2-b(w)),
for p(w) the packet drop rate for congestion window w. From for p(w) the packet drop rate for congestion window w. From
inverting Equation (1), we get p(w) as follows: inverting Equation (1), we get p(w) as follows:
p(w) = 0.078/w^1.2. p(w) = 0.078/w^1.2.
We assume that experimental implementations of HighSpeed TCP for We assume that experimental implementations of HighSpeed TCP for
further investigation will use a pre-computed look-up table for further investigation will use a pre-computed look-up table for
finding a(w) and b(w). For example, the implementation from Tom finding a(w) and b(w). For example, the implementation from Tom
Dunigan adjusts the a(w) and b(w) parameters every 0.1 seconds. In Dunigan adjusts the a(w) and b(w) parameters every 0.1 seconds. In
the appendix we give such a table for our default values of the appendix we give such a table for our default values of
Low_Window = 38, High_Window = 83,000, High_P = 10^-7, and Low_Window = 38, High_Window = 83,000, High_P = 10^-7, and
High_Decrease = 0.1. These are also the default values in the NS High_Decrease = 0.1. These are also the default values in the NS
simulator; example simulations in NS can be run with the command simulator; example simulations in NS can be run with the command
"./test-all-tcpHighspeed" in the directory tcl/test. "./test-all-tcpHighspeed" in the directory tcl/test.
8. An alternate, linear response functions. 8. An alternate, linear response functions.
In this section we explore an alternate, linear response function for In this section we explore an alternate, linear response function
HighSpeed TCP that has been proposed by a number of other people, in for HighSpeed TCP that has been proposed by a number of other
particular by Glenn Vinnicombe and Tom Kelly. Similarly, it has been people, in particular by Glenn Vinnicombe and Tom Kelly. Similarly,
suggested by others that a less "ad-hoc" guideline for a response it has been suggested by others that a less "ad-hoc" guideline for a
function for HighSpeed TCP would be to specify a constant value for response function for HighSpeed TCP would be to specify a constant
the number of round-trip times between congestion events. value for the number of round-trip times between congestion events.
Assume that we keep the value of Low_Window as 38 MSS-sized segments, Assume that we keep the value of Low_Window as 38 MSS-sized
indicating when the HighSpeed response function diverges from the segments, indicating when the HighSpeed response function diverges
current TCP response function, but that we modify the High_Window and from the current TCP response function, but that we modify the
High_P parameters that specify the upper range of the HighSpeed High_Window and High_P parameters that specify the upper range of
response function. In particular, consider the response function the HighSpeed response function. In particular, consider the
given by High_Window = 380,000 and High_P = 10^-7, with Low_Window = response function given by High_Window = 380,000 and High_P = 10^-7,
38 and Low_P = 10^-3 as before. with Low_Window = 38 and Low_P = 10^-3 as before.
Using the equations in Section 5, this would give the following Using the equations in Section 5, this would give the following
Linear response function, for w > Low_Window: Linear response function, for w > Low_Window:
W = 0.038/p. W = 0.038/p.
This Linear HighSpeed response function is illustrated in Table 7 This Linear HighSpeed response function is illustrated in Table 7
below. For HighSpeed TCP, the number of round-trip times between below. For HighSpeed TCP, the number of round-trip times between
losses, 1/(pW), equals 1/0.38, or equivalently, 26, for W > 38 losses, 1/(pW), equals 1/0.38, or equivalently, 26, for W > 38
segments. segments.
Packet Drop Rate P Congestion Window W RTTs Between Losses Packet Drop Rate P Congestion Window W RTTs Between Losses
------------------ ------------------- ------------------- ------------------ ------------------- -------------------
10^-2 12 8 10^-2 12 8
10^-3 38 26 10^-3 38 26
10^-4 380 26 10^-4 380 26
10^-5 3800 26 10^-5 3800 26
10^-6 38000 26 10^-6 38000 26
10^-7 380000 26 10^-7 380000 26
10^-8 3800000 26 10^-8 3800000 26
10^-9 38000000 26 10^-9 38000000 26
10^-10 380000000 26 10^-10 380000000 26
Table 7: An Alternate, Linear TCP Response Function for HighSpeed Table 7: An Alternate, Linear TCP Response Function for HighSpeed
TCP. The average congestion window W in MSS-sized segments is given TCP. The average congestion window W in MSS-sized segments is given
as a function of the packet drop rate P. as a function of the packet drop rate P.
Given a constant decrease b(w) of 1/2, this would give an increase Given a constant decrease b(w) of 1/2, this would give an increase
a(w) of w/Low_Window, or equivalently, an constant increase of a(w) of w/Low_Window, or equivalently, a constant increase of
1/Low_Window packets per acknowledgement, for w > Low_Window. 1/Low_Window packets per acknowledgement, for w > Low_Window.
Another possibility is Scalable TCP [K03], which uses a fixed Another possibility is Scalable TCP [K03], which uses a fixed
decrease b(w) of 1/8 and a fixed increase per acknowledgement of decrease b(w) of 1/8 and a fixed increase per acknowledgement of
0.01. This gives an increase a(w) per window of 0.005 w, for a TCP 0.01. This gives an increase a(w) per window of 0.005 w, for a TCP
with delayed acknowledgements. with delayed acknowledgements, for pure MIMD.
The relative fairness between the alternate Linear response function The relative fairness between the alternate Linear response function
and the standard TCP response function is illustrated below in Table and the standard TCP response function is illustrated below in Table
8. 8.
Packet Drop Rate P Fairness Aggregate Window Bandwidth Packet Drop Rate P Fairness Aggregate Window Bandwidth
------------------ -------- ---------------- --------- ------------------ -------- ---------------- ---------
10^-2 1.0 24 2.8 Mbps 10^-2 1.0 24 2.8 Mbps
10^-3 1.0 76 9.1 Mbps 10^-3 1.0 76 9.1 Mbps
10^-4 3.2 500 60.0 Mbps 10^-4 3.2 500 60.0 Mbps
10^-5 15.1 4179 501.4 Mbps 10^-5 15.1 4179 501.4 Mbps
10^-6 31.6 39200 4.7 Gbps 10^-6 31.6 39200 4.7 Gbps
10^-7 100.1 383795 46.0 Gbps 10^-7 100.1 383795 46.0 Gbps
10^-8 316.6 3812000 457.4 Gbps
10^-9 1001.3 38037948 4564.5 Gbps
10^-10 3166.6 380120000 45614.4 Gbps
Table 8: Relative Fairness between the Linear HighSpeed and Standard Table 8: Relative Fairness between the Linear HighSpeed and Standard
Response Functions. Response Functions.
One attraction of the linear response function is that it is scale- One attraction of the linear response function is that it is scale-
invariant, with a fixed increase in the congestion window per invariant, with a fixed increase in the congestion window per
acknowledgement, and a fixed number of round-trip times between loss acknowledgement, and a fixed number of round-trip times between loss
events. My own assumption would be that having a fixed length for events. My own assumption would be that having a fixed length for
the congestion epoch in round-trip times, regardless of the packet the congestion epoch in round-trip times, regardless of the packet
drop rate, would be a poor fit for an imprecise and imperfect world drop rate, would be a poor fit for an imprecise and imperfect world
with routers with a range of queue management mechanisms, such as the with routers with a range of queue management mechanisms, such as
Drop-Tail queue management that is common today. For example, a the Drop-Tail queue management that is common today. For example, a
response function with a fixed length for the congestion epoch in response function with a fixed length for the congestion epoch in
round-trip times might give less clearly-differentiated feedback in round-trip times might give less clearly-differentiated feedback in
an environment with steady-state background losses at fixed intervals an environment with steady-state background losses at fixed
for all flows (as might occur with a wireless link with occasional intervals for all flows (as might occur with a wireless link with
short error bursts, giving losses for all flows every N seconds occasional short error bursts, giving losses for all flows every N
regardless of their sending rate). seconds regardless of their sending rate).
While it is not a goal to have perfect fairness in an environment While it is not a goal to have perfect fairness in an environment
with synchronized losses, it would be good to have moderately with synchronized losses, it would be good to have moderately
acceptable performance in this regime. This goal might argue against acceptable performance in this regime. This goal might argue
a response function with a constant number of round-trip times against a response function with a constant number of round-trip
between congestion events. However, this is a question that could times between congestion events. However, this is a question that
clearly use additional research and investigation. In addition, could clearly use additional research and investigation. In
flows with different round-trip times would have different time addition, flows with different round-trip times would have different
durations for congestion epochs even in the model with a linear time durations for congestion epochs even in the model with a linear
response function. response function.
The third column of Table 8, the Aggregate Window, gives the The third column of Table 8, the Aggregate Window, gives the
aggregate congestion window of two competing TCP connections, one aggregate congestion window of two competing TCP connections, one
with Linear HighSpeed TCP and one with Standard TCP, given the packet with Linear HighSpeed TCP and one with Standard TCP, given the
drop rate specified in the first column. From Table 8, a Linear packet drop rate specified in the first column. From Table 8, a
HighSpeed TCP connection would receive fifteen times the bandwidth of Linear HighSpeed TCP connection would receive fifteen times the
a Standard TCP in an environment with a packet drop rate of 10^-5. bandwidth of a Standard TCP in an environment with a packet drop
This would occur when the two flows sharing a single pipe achieved an rate of 10^-5. This would occur when the two flows sharing a single
aggregate window of 4179 packets. Given a round-trip time of 100 ms pipe achieved an aggregate window of 4179 packets. Given a round-
and a packet size of 1500 bytes, this would occur with an available trip time of 100 ms and a packet size of 1500 bytes, this would
bandwidth for the two competing flows of 501 Mbps. Thus, because the occur with an available bandwidth for the two competing flows of 501
Linear HighSpeed TCP is more aggressive than the HighSpeed TCP Mbps. Thus, because the Linear HighSpeed TCP is more aggressive
proposed above, it also is less fair when competing with Standard TCP than the HighSpeed TCP proposed above, it also is less fair when
in a high-bandwidth environment. competing with Standard TCP in a high-bandwidth environment.
9. Tradeoffs for Choosing Congestion Control Parameters. 9. Tradeoffs for Choosing Congestion Control Parameters.
A range of metrics can be used for evaluating choices for congestion A range of metrics can be used for evaluating choices for congestion
control parameters for HighSpeed TCP. My assumption in this section control parameters for HighSpeed TCP. My assumption in this section
is that for a response function of the form w = c/p^d, for constant c is that for a response function of the form w = c/p^d, for constant
and exponent d, the only response functions that would be considered c and exponent d, the only response functions that would be
are response functions with 1/2 <= d <= 1. The two ends of this considered are response functions with 1/2 <= d <= 1. The two ends
spectrum are represented by current TCP, with d = 1/2, and by the of this spectrum are represented by current TCP, with d = 1/2, and
linear response function described in Section 8 above, with d = 1. by the linear response function described in Section 8 above, with d
HighSpeed TCP lies somewhere in the middle of the spectrum, with d = = 1. HighSpeed TCP lies somewhere in the middle of the spectrum,
0.835. with d = 0.835.
Response functions with exponents less than 1/2 can be eliminated Response functions with exponents less than 1/2 can be eliminated
from consideration because they would be even worse than standard TCP from consideration because they would be even worse than standard
in accomodating connections with high congestion windows. TCP in accomodating connections with high congestion windows.
9.1. The Number of Round-Trip Times between Loss Events. 9.1. The Number of Round-Trip Times between Loss Events.
Response functions with exponents greater than 1 can be eliminated Response functions with exponents greater than 1 can be eliminated
from consideration because for these response functions, the number from consideration because for these response functions, the number
of round-trip times between loss events decreases as congestion of round-trip times between loss events decreases as congestion
decreases. For a response function of w = c/p^d, with one loss event decreases. For a response function of w = c/p^d, with one loss
or congestion event every 1/p packets, the number of round-trip times event or congestion event every 1/p packets, the number of round-
between loss events is w^((1/d)-1)/c^(1/d). Thus, for standard TCP trip times between loss events is w^((1/d)-1)/c^(1/d). Thus, for
the number of round-trip times between loss events is linear in w. standard TCP the number of round-trip times between loss events is
In contrast, one attraction of the linear response function, as linear in w. In contrast, one attraction of the linear response
described in Section 8 above, is that it is scale-invariant, in terms function, as described in Section 8 above, is that it is scale-
of a fixed increase in the congestion window per acknowledgement, and invariant, in terms of a fixed increase in the congestion window per
a fixed number of round-trip times between loss events. acknowledgement, and a fixed number of round-trip times between loss
events.
However, for a response function with d > 1, the number of round-trip However, for a response function with d > 1, the number of round-
times between loss events would be proportional to w^((1/d)-1), for a trip times between loss events would be proportional to w^((1/d)-1),
negative exponent ((1/d)-1), setting smaller as w increases. This for a negative exponent ((1/d)-1), setting smaller as w increases.
would seem undesirable. This would seem undesirable.
9.2. The Number of Packet Drops per Loss Event, with Drop-Tail. 9.2. The Number of Packet Drops per Loss Event, with Drop-Tail.
A TCP connection increases its sending rate by a(w) packets per A TCP connection increases its sending rate by a(w) packets per
round-trip time, and in a Drop-Tail environment, this is likely to round-trip time, and in a Drop-Tail environment, this is likely to
result in a(w) dropped packets during a single loss event. One result in a(w) dropped packets during a single loss event. One
attraction of standard TCP is that it has a fixed increase per round- attraction of standard TCP is that it has a fixed increase per
trip time of one packet, minimizing the number of packets that would round-trip time of one packet, minimizing the number of packets that
be dropped in a Drop-Tail environment. For an environment with some would be dropped in a Drop-Tail environment. For an environment
form of Active Queue Management, and in particular for an environment with some form of Active Queue Management, and in particular for an
that uses ECN, the number of packets dropped in a single congestion environment that uses ECN, the number of packets dropped in a single
event would not be a problem. However, even in these environments, congestion event would not be a problem. However, even in these
larger increases in the sending rate per round-trip time result in environments, larger increases in the sending rate per round-trip
larger stresses on the ability of the queues in the router to absorb time result in larger stresses on the ability of the queues in the
the fluctuations. router to absorb the fluctuations.
HighSpeed TCP plays a middle ground between the metrics of a moderate HighSpeed TCP plays a middle ground between the metrics of a
number of round-trip times between loss events, and a moderate moderate number of round-trip times between loss events, and a
increase in the sending rate per round-trip time. As shown in moderate increase in the sending rate per round-trip time. As shown
Appendix B, for a congestion window of 83,000 packets, HighSpeed TCP in Appendix B, for a congestion window of 83,000 packets, HighSpeed
increases its sending rate by 70 packets per round-trip time, TCP increases its sending rate by 70 packets per round-trip time,
resulting in roughly 70 packet drops for each congestion event in a resulting in at most 70 packet drops when the buffer overflows in a
Drop-Tail environment. This increased aggressiveness is the price Drop-Tail environment. This increased aggressiveness is the price
paid by HighSpeed TCP for its increased scalability. A large number paid by HighSpeed TCP for its increased scalability. A large number
of packets dropped per congestion event could result in synchronized of packets dropped per congestion event could result in synchronized
drops from multiple flows, with a possible loss of throughput as a drops from multiple flows, with a possible loss of throughput as a
result. result.
Scalable TCP has an increase a(w) of 0.005 w packets per round-trip Scalable TCP has an increase a(w) of 0.005 w packets per round-trip
time. For a congestion window of 83,000 packets, this gives an time. For a congestion window of 83,000 packets, this gives an
increase of 415 packets per round-trip time, resulting in roughly 415 increase of 415 packets per round-trip time, resulting in roughly
packet drops per congestion event in a Drop-Tail environment. 415 packet drops per congestion event in a Drop-Tail environment.
Thus, HighSpeed TCP and its variants place increased demands on queue Thus, HighSpeed TCP and its variants place increased demands on
management in routers, relative to Standard TCP. (This is rather queue management in routers, relative to Standard TCP. (This is
similar to the increased demands on queue management that would rather similar to the increased demands on queue management that
result from using N parallel TCP connections instead of a single would result from using N parallel TCP connections instead of a
Standard TCP connection.) single Standard TCP connection.)
10. Slow-Start. 10. Related Issues
An companion internet-draft on "Limited Slow-Start for TCP with Large 10.1. Slow-Start.
Congestion Windows" [F02b] proposes a modification to TCP's slow-
start procedure that can significantly improve the performance of TCP
connections slow-starting up to large congestion windows. For TCP
connections that are able to use congestion windows of thousands (or
tens of thousands) of MSS-sized segments (for MSS the sender's
MAXIMUM SEGMENT SIZE), the current slow-start procedure can result in
increasing the congestion window by thousands of segments in a single
round-trip time. Such an increase can easily result in thousands of
packets being dropped in one round-trip time. This is often counter-
productive for the TCP flow itself, and is also hard on the rest of
the traffic sharing the congested link.
[F02b] proposes Limited Slow-Start, limiting the number of segments An companion internet-draft on "Limited Slow-Start for TCP with
by which the congestion window is increased for one window of data Large Congestion Windows" [F02b] proposes a modification to TCP's
during slow-start, in order to improve performance for TCP slow-start procedure that can significantly improve the performance
connections with large congestion windows. We have separated out of TCP connections slow-starting up to large congestion windows.
Limited Slow-Start to a separate draft because it can be used both For TCP connections that are able to use congestion windows of
with Standard or with HighSpeed TCP. thousands (or tens of thousands) of MSS-sized segments (for MSS the
sender's MAXIMUM SEGMENT SIZE), the current slow-start procedure can
result in increasing the congestion window by thousands of segments
in a single round-trip time. Such an increase can easily result in
thousands of packets being dropped in one round-trip time. This is
often counter-productive for the TCP flow itself, and is also hard
on the rest of the traffic sharing the congested link.
Limited Slow-Start is illustrated in the NS simulator, for snapshots [F02b] proposes Limited Slow-Start, limiting the number of segments
after May 1, 2002, in the tests "./test-all-tcpHighspeed tcp1A" and by which the congestion window is increased for one window of data
"./test-all-tcpHighspeed tcpHighspeed1" in the subdirectory during slow-start, in order to improve performance for TCP
"tcl/lib". connections with large congestion windows. We have separated out
Limited Slow-Start to a separate draft because it can be used both
with Standard or with HighSpeed TCP.
In order for best-effort flows to safely start-up faster than slow- Limited Slow-Start is illustrated in the NS simulator, for snapshots
start, e.g., in future high-bandwidth networks, we believe that it after May 1, 2002, in the tests "./test-all-tcpHighspeed tcp1A" and
would be necessary for the flow to have explicit feedback from the "./test-all-tcpHighspeed tcpHighspeed1" in the subdirectory
routers along the path. There are a number of proposals for this, "tcl/lib".
ranging from a minimal proposal for an IP option that allows TCP SYN
packets to collect information from routers along the path about the
allowed initial sending rate [J02], to proposals with more power that
require more fine-tuned and continuous feedback from routers. These
proposals all are somewhat longer-term proposals that the HighSpeed
TCP proposal in this document, requiring longer lead times and more
coordination for deployment, and will be discussed in later
documents.
11. Other limitations on window size. In order for best-effort flows to safely start-up faster than slow-
start, e.g., in future high-bandwidth networks, we believe that it
would be necessary for the flow to have explicit feedback from the
routers along the path. There are a number of proposals for this,
ranging from a minimal proposal for an IP option that allows TCP SYN
packets to collect information from routers along the path about the
allowed initial sending rate [J02], to proposals with more power
that require more fine-tuned and continuous feedback from routers.
These proposals all are somewhat longer-term proposals than the
HighSpeed TCP proposal in this document, requiring longer lead times
and more coordination for deployment, and will be discussed in later
documents.
The TCP header uses a 16-bit field to report the receive window size 10.2. Limiting burstiness on short time scales.
to the sender. Unmodified, this allows a window size of at most
2**16 = 65K bytes. With window scaling, the maximum window size is Because the congestion window achieved by a HighSpeed TCP connection
2**30 = 1073M bytes [RFC 1323]. Given 1500-byte packets, this allows could be quite large, there is a possibility for the sender to send
a window of up to 715,000 packets. a large burst of packets in response to a single acknowledgement.
This could happen, for example, when there is congestion or
reordering on the reverse path, and the sender receives an
acknowledgement acknowledging hundreds or thousands of new packets.
Such a burst would also result if the application was idle for a
short period of time less than a round-trip time, and then suddenly
had lots of data available to send. In this case, it would be
useful for the HighSpeed TCP connection to have some method for
limiting bursts.
We do not in this document specify TCP mechanisms for reducing the
short-term burstiness. One possible mechanism is to use some form
of rate-based pacing, and another possibility is to use maxburst,
which limits the number of packets that are sent in response to a
single acknowledgement. We would caution, however, against a
permanent reduction in the congestion window as a mechanism for
limiting short-term bursts. Such a mechanism has been deployed in
some TCP stacks, and our view would be that using permanent
reductions of the congestion window to reduce transient bursts would
be a bad idea [Fl03].
10.3. Other limitations on window size.
The TCP header uses a 16-bit field to report the receive window size
to the sender. Unmodified, this allows a window size of at most
2**16 = 65K bytes. With window scaling, the maximum window size is
2**30 = 1073M bytes [RFC 1323]. Given 1500-byte packets, this
allows a window of up to 715,000 packets.
10.4. Implementation issues.
One implementation issue that has been raised with HighSpeed TCP is
that with congestion windows of 4MB or more, the handling of
successive SACK packets after a packet is dropped becomes very time-
consuming at the TCP sender [S03]. Tom Kelly's Scalable TCP
includes a "SACK Fast Path" patch that addresses this problem.
The issues addressed in the Web100 project, the Net100 project, and
related projects about the tuning necessary to achieve high
bandwidth data rates with TCP apply to HighSpeed TCP as well
[Net100, Web100].
11. Deployment issues.
11.1. Deployment issues of HighSpeed TCP
We do not claim that the HighSpeed TCP modification to TCP described
in this paper is an optimal transport protocol for high-bandwidth
environments. Based on our experiences with HighSpeed TCP in the NS
simulator [NS], on simulation studies [SA03], and on experimental
reports [ABLLS03,D02,CC03,F03], we believe that HighSpeed TCP
improves the performance of TCP in high-bandwidth environments, and
we are documenting it for the benefit of the IETF community. We
encourage the use of HighSpeed TCP, and of its underlying response
function, and we further encourage feedback about operational
experiences with this or related modifications.
We note that in environments typical of much of the current
Internet, HighSpeed TCP behaves exactly as does Standard TCP today.
This is the case any time the congestion window is less than 38
segments.
Bandwidth Avg Cwnd w (pkts) Increase a(w) Decrease b(w)
--------- ----------------- ------------- -------------
1.5 Mbps 12.5 1 0.50
10 Mbps 83 1 0.50
100 Mbps 833 6 0.35
1 Gbps 8333 26 0.22
10 Gbps 83333 70 0.10
Table 9: Performance of a HighSpeed TCP connection.
To help calibrate, Table 9 considers a TCP connection with 1500-byte
packets, an RTT of 100 ms (including average queueing delay), and no
competing traffic, and shows the average congestion window if that
TCP connection had a pipe all to itself and fully used the link
bandwidth, for a range of bandwidths for the pipe. This assumes
that the TCP connection would use Table 12 in determining its
increase and decrease parameters. The first column of Table 9 gives
the bandwidth, and the second column gives the average congestion
window w needed to utilize that bandwidth. The third column show
the increase a(w) in segments per RTT for window w. The fourth
column show the decrease b(w) for that window w (where the TCP
sender decreases the congestion window from w to w(1-b(w)) segments
after a loss event). We note that the actual congestion window when
a loss occurs is likely to be greater than the average congestion
window w in column 2, so the decrease parameter used could be
slightly smaller than the one given in column 4 of Table 9.
Table 9 shows that a HighSpeed TCP over a 10 Mbps link behaves
exactly the same as a Standard TCP connection, even in the absence
of competing traffic. One can think of the congestion window
staying generally in the range of 55 to 110 segments, with the
HighSpeed TCP behavior being exactly the same as the behavior of
Standard TCP. (If the congestion window is ever 128 segments or
more, then the HighSpeed TCP increases by two segments per RTT
instead of by one, and uses a decrease parameter of 0.44 instead of
0.50.)
Table 9 shows that for a HighSpeed TCP connection over a 100 Mbps
link, with no competing traffic, HighSpeed TCP behaves roughly as
aggressively as six parallel TCP connections, increasing its
congestion window by roughly six segments per round-trip time, and
with a decrease parameter of roughly 1/3 (corresponding to
decreasing down to 2/3-rds of its old congestion window, rather than
to half, in response to a loss event).
For a Standard TCP connection in this environment, the congestion
window could be thought of as varying generally in the range of 550
to 1100 segments, with an average packet drop rate of 2.2 * 10^-6
(corresponding to a bit error rate of 1.8 * 10^-10), or
equivalently, roughly 55 seconds between congestion events. While a
Standard TCP connection could sustain such a low packet drop rate in
a carefully controlled environment with minimal competing traffic,
we would contend that in an uncontrolled best-effort environment
with even a small amount of competing traffic, the occasional
congestion events from smaller competing flows could easily be
sufficient to prevent a Standard TCP flow with no lower-speed
bottlenecks from fully utilizing the available bandwidth of the
underutilized 100 Mbps link.
That is, we would content that in the environment of 100 Mbps links
with a significant amount of available bandwidth, Standard TCP would
sometimes be unable to fully utilize the link bandwidth, and that
HighSpeed TCP would be an improvement in this regard. We would
further contend that in this environment, the behavior of HighSpeed
TCP is sufficiently close to that of Standard TCP that HighSpeed TCP
would be safe to deploy in the current Internet.
11.2. Deployment issues of Scalable TCP
We believe that Scalable TCP and HighSpeed TCP have sufficiently
similar response functions that they could easily coexist in the
Internet. However, we have not investigated Scalable TCP
sufficiently to be able to claim, in this document, that Scalable
TCP is safe for a widespread deployment in the current Internet.
Bandwidth Avg Cwnd w (pkts) Increase a(w) Decrease b(w)
--------- ----------------- ------------- -------------
1.5 Mbps 12.5 1 0.50
10 Mbps 83 0.4 0.125
100 Mbps 833 4.1 0.125
1 Gbps 8333 41.6 0.125
10 Gbps 83333 416.5 0.125
Table 10: Performance of a Scalable TCP connection.
Table 10 shows the performance of a Scalable TCP connection with
1500-byte packets, an RTT of 100 ms (including average queueing
delay), and no competing traffic. The TCP connection is assumed to
use delayed acknowledgements. The first column of Table 10 gives
the bandwidth, the second column gives the average congestion window
needed to utilize that bandwidth, and the third and fourth columns
give the increase and decrease parameters.
Note that even in an environment with a 10 Mbps link, Scalable TCP's
behavior is considerably different from that of Standard TCP. The
increase parameter is smaller than that of Standard TCP, and the
decrease is smaller also, 1/8-th instead of 1/2. That is, for 10
Mbps links, Scalable TCP increases less aggressively than Standard
TCP or HighSpeed TCP, but decreases less aggressively as well.
In an environment with a 100 Mbps link, Scalable TCP has an increase
parameter of roughly four segments per round-trip time, with the
same decrease parameter of 1/8-th. A comparison of Tables 9 and 10
shows that for this scenario of 100 Mbps links, HighSpeed TCP
increases more aggressively than Scalable TCP.
Next we consider the relative fairness between Standard TCP,
HighSpeed TCP and Scalable TCP. The relative fairness between
HighSpeed TCP and Standard TCP was shown in Table 5 earlier in this
document, and the relative fairness between Scalable TCP and
Standard TCP was shown in Table 8. Following the approach in
Section 6, for a given packet drop rate p, for p < 10^-3, we can
estimate the relative fairness between Scalable and HighSpeed TCP as
W_Scalable/W_HighSpeed. This relative fairness is shown in Table 11
below. The bandwidth in the last column of Table 11 is the
aggregate bandwidth of the two competing flows given 100 ms round-
trip times and 1500-byte packets.
Packet Drop Rate P Fairness Aggregate Window Bandwidth
------------------ -------- ---------------- ---------
10^-2 1.0 24 2.8 Mbps
10^-3 1.0 76 9.1 Mbps
10^-4 1.4 643 77.1 Mbps
10^-5 2.1 5595 671.4 Mbps
10^-6 3.1 50279 6.0 Gbps
10^-7 4.5 463981 55.7 Gbps
Table 11: Relative Fairness between the Scalable and HighSpeed
Response Functions.
The second row of Table 11 shows that for a Scalable TCP and a
HighSpeed TCP flow competing in an environment with 100 ms RTTs and
a 10 Mbps pipe, the two flows would receive essentially the same
bandwidth. The next row shows that for a Scalable TCP and a
HighSpeed TCP flow competing in an environment with 100 ms RTTs and
a 100 Mbps pipe, the Scalable TCP flow would receive roughly 50%
more bandwidth than would HighSpeed TCP. Table 11 shows the
relative fairness in higher-bandwidth environments as well. This
relative fairness seems sufficient that there should be no problems
with Scalable TCP and HighSpeed TCP coexisting in the same
environment as Experimental variants of TCP.
We note that one question that requires more investigation with
Scalable TCP is that of convergence to fairness in environments with
Drop-Tail queue management.
12. Related Work in HighSpeed TCP. 12. Related Work in HighSpeed TCP.
HighSpeed TCP has been separately investigated in simulations by HighSpeed TCP has been separately investigated in simulations by
Sylvia Ratnasamy and by Evandro de Souza, and reports of some of Sylvia Ratnasamy and by Evandro de Souza [SA03]. The simulations in
these simulations should be available shortly. The simulations by [SA03] verify the fairness properties of HighSpeed TCP when sharing
Evandro verify the fairness properties of HighSpeed TCP when sharing a link with Standard TCP.
a link with Standard TCP.
These simulations explore the relative fairness of HighSpeed TCP These simulations explore the relative fairness of HighSpeed TCP
flows when competing with Standard TCP. The simulation environment flows when competing with Standard TCP. The simulation environment
include background forward and reverse-path TCP traffic limited by includes background forward and reverse-path TCP traffic limited by
the TCP receive window, along with a small amount of forward and the TCP receive window, along with a small amount of forward and
reverse-path traffic from the web traffic generator. Most of the reverse-path traffic from the web traffic generator. Most of the
simulations so far explore performance on a simple dumbbell topology simulations so far explore performance on a simple dumbbell topology
with a 1Gbps link with a propagation delay of 50 ms. Simulations with a 1 Gbps link with a propagation delay of 50 ms. Simulations
have been run both the Adaptive RED and with DropTail queue have been run with Adaptive RED and with DropTail queue management.
management.
Future work to explore in more detail includes convergence times The simulations in [SA03] explore performance with a varying number
after new flows start-up; recovery time after a transient outage; the of competing flows, with the competing traffic being all standard
response to sudden severe congestion, and investigations of the TCP; all HighSpeed TCP; or a mix of standard and HighSpeed TCP. For
potential for oscillations. Additional future work includes the simulations in [SA03] with RED queue management, the relative
evaluating more fully the choices of parameters for HighSpeed TCP. fairness between standard and HighSpeed TCP is consistent with the
We invite contributions from others in this work. relative fairness predicted in Table 5. For the simulations with
Drop Tail queues, the relative fairness is more skewed, with the
HighSpeed TCP flows receiving an even larger share of the link
bandwidth. This is not surprising; with Active Queue Management at
the congested link, the fraction of packet drops received by each
flow should be roughly proportional to that flow's share of the link
bandwidth, while this property no longer holds with Drop Tail queue
management. We also note that relative fairness in simulations with
Drop Tail queue management can sometimes depend on small details of
the simulation scenario, and that Drop Tail simulations need special
care to avoid phase effects [F92].
Suggestions to other citations of related work would also be welcome. [SA03] explores the bandwidth `stolen' by HighSpeed TCP from
standard TCP by exploring the fraction of the link bandwidth N
standard TCP flows receive when competing against N other standard
TCP flows, and comparing this to the fraction of the link bandwidth
the N standard TCP flows receive when competing against N HighSpeed
TCP flows. For the 1 Gbps simulation scenarios dominated by long-
lived traffic, a small number of standard TCP flows are able to
achieve high link utilization, and the HighSpeed TCP flows can be
viewed as stealing bandwidth from the competing standard TCP flows,
as predicted in Section 6 on the Fairness Implications of the
HighSpeed Response Function. However, [SA03] shows that when even a
small fraction of the link bandwidth is used by more bursty, short
TCP connections, the standard TCP flows are unable to achieve high
link utilization, and the HighSpeed TCP flows in this case are not
`stealing' bandwidth from the standard TCP flows, but instead are
using bandwidth that otherwise would not be utilized.
13. Relationship to other Work. The conclusions of [SA03] are that "HighSpeed TCP behaved as forseen
by its response function, and appears to be a real and viable option
for use on high-speed wide area TCP connections."
Our assumption is that HighSpeed TCP will be used along with the TCP Future work that could be explored in more detail includes
SACK option, and also with the increased Initial Window of three or convergence times after new flows start-up; recovery time after a
four segments, as allowed by [AFP02]. For paths that have transient outage; the response to sudden severe congestion, and
substantial reordering, TCP performance would be greatly improved by investigations of the potential for oscillations. We invite
some of the mechanisms still in the research stages for robust contributions from others in this work.
performance in the presence of reordered packets.
Our view is that HighSpeed TCP is largely orthogonal to proposals for 13. Relationship to other Work.
higher PMTU (Path MTU) values [M02]. Unlike changes to the PMTU,
HighSpeed TCP does not require any changes in the network or at the
TCP receiver, and works well in the current Internet. Our assumption
is that HighSpeed TCP would be useful even with larger values for the
PMTU. In particular, unlike the current congestion window, the PMTU
gives no information about the bandwidth-delay product available to
that particular flow.
A related approach is that of a virtual MTU, where the actual MTU of Our assumption is that HighSpeed TCP will be used with the TCP SACK
the path might be limited [VMSS,S02]. The virtual MTU approach has option, and also with the increased Initial Window of three or four
not been fully investigated, and we do not explore the virtual MTU segments, as allowed by [RFC3390]. For paths that have substantial
approach further in this document. reordering, TCP performance would be greatly improved by some of the
mechanisms still in the research stages for robust performance in
the presence of reordered packets.
14. Conclusions. Our view is that HighSpeed TCP is largely orthogonal to proposals
for higher PMTU (Path MTU) values [M02]. Unlike changes to the
PMTU, HighSpeed TCP does not require any changes in the network or
at the TCP receiver, and works well in the current Internet. Our
assumption is that HighSpeed TCP would be useful even with larger
values for the PMTU. Unlike the current congestion window, the PMTU
gives no information about the bandwidth-delay product available to
that particular flow.
This is an initial proposal, and we are asking from feedback from the A related approach is that of a virtual MTU, where the actual MTU of
wider community. We have explored this proposal in simulations, the path might be limited [VMSS,S02]. The virtual MTU approach has
though we have not yet finished our reports on these simulations. We not been fully investigated, and we do not explore the virtual MTU
would welcome additional analysis, simulations, and particularly, approach further in this document.
experimentation. More information on simuations and experiments is
available from the HighSpeed TCP Web Page [HSTCP].
There are three parameters that determine the HighSpeed Response 14. Conclusions.
Function, and an additional parameter that determines HighSpeed TCP's
tradeoffs between increases and decreases using that response
function. We solicit feedback on our setting of these parameters as
well as on other issues.
We are bringing this proposal to the IETF to be considered as an This document has proposed HighSpeed TCP, a modification to TCP's
Experimental RFC. One reason to bring this to the IETF at this stage congestion control mechanism for use with TCP connections with large
is that HighSpeed TCP proposes a rather significant change in the congestion windows. We have explored this proposal in simulations,
underlying TCP response function, and in our view any such change and others have explored HighSpeed TCP with experiments, and we
would have to be globally agreed-upon. It seems advisable to us to believe HighSpeed TCP to be safe to deploy on the current Internet.
bring such a proposal to the IETF for feedback even in its We would welcome additional analysis, simulations, and particularly,
preliminary stages. experimentation. More information on simuations and experiments is
available from the HighSpeed TCP Web Page [HSTCP]. There are
several independent implementations of HighSpeed TCP [D02,F03] and
of Scalable TCP [K03] for further investigation.
Another reason to bring this proposal to the IETF is that, while We are bringing this proposal to the IETF to be considered as an
several people have conducted evaluations of HighSpeed TCP using Experimental RFC.
simulations, our belief is that the "real" evaluations will have to
happen in experiments and in actual deployment. As part of this
experimentation, HighSpeed TCP has been implemented in the Linux
2.4.16 Web100 kernel [HSTCP]. It seemed to us that it was advisable,
at this stage, to bring the proposal for HighSpeed TCP to the IETF
and to seek Experimental status.
15. Acknowledgements 15. Acknowledgements
The HighSpeed TCP proposal is from joint work with Sylvia Ratnasamy The HighSpeed TCP proposal is from joint work with Sylvia Ratnasamy
and Scott Shenker (and was initiated by Scott Shenker). Additional and Scott Shenker (and was initiated by Scott Shenker). Additional
investigations of HighSpeed TCP were joint work with Evandro de Souza investigations of HighSpeed TCP were joint work with Evandro de
and Deb Agarwal. We thank Tom Dunigan for the implementation in the Souza and Deb Agarwal. We thank Tom Dunigan for the implementation
Linux 2.4.16 Web100 kernel, and for resulting experimentation with in the Linux 2.4.16 Web100 kernel, and for resulting experimentation
HighSpeed TCP. We are grateful to the End-to-End Research Group, the with HighSpeed TCP. We are grateful to the End-to-End Research
members of the Transport Area Working Group, and to members of the Group, the members of the Transport Area Working Group, and to
IPAM program in Large Scale Communication Networks for feedback. We members of the IPAM program in Large Scale Communication Networks
thank Glenn Vinnicombe for framing the Linear response function in for feedback. We thank Glenn Vinnicombe for framing the Linear
the parameters of HighSpeed TCP. We are also grateful for response function in the parameters of HighSpeed TCP. We are also
contributions and feedback from the following individuals: Tom Kelly, grateful for contributions and feedback from the following
Jitendra Padhye, Stanislav Shalunov, Paul Sutter, Brian Tierney, Joe individuals: Les Cottrell, Mitchell Erblich, Jeffrey Hsu, Tom Kelly,
Touch. Thanks to Jeffrey Hsu and Andrew Reiter for feedback on Jitendra Padhye, Andrew Reiter, Stanislav Shalunov, Alex Solan, Paul
earlier versions of this document. Sutter, Brian Tierney, Joe Touch.
16. Normative References 16. Normative References
[RFC2581] M. Allman and V. Paxson, "TCP Congestion Control", RFC [RFC2581] M. Allman, V. Paxson, and W. Stevens, "TCP Congestion
2581, April 1999. Control", RFC 2581, April 1999.
17. Informative References 17. Informative References
[AFP02] Allman, M., Floyd, S., and Partridge, C., "Increasing TCP's [ABLLS03] A. Antony, J. Blom, C. de Laat, J. Lee, and W. Sjouw,
Initial Window", internet-draft draft-ietf-tsvwg-initwin-04.txt, Macroscopic Examination of TCP Flows over Transatlantic Links,
work-in-progress, June 2002. January 2003. URL
"http://carol.wins.uva.nl/%7Edelaat/techrep-2003-2-tcp.pdf".
[BBFS01] Deepak Bansal, Hari Balakrishnan, Sally Floyd, and Scott [BBFS01] Deepak Bansal, Hari Balakrishnan, Sally Floyd, and Scott
Shenker, "Dynamic Behavior of Slowly-Responsive Congestion Control Shenker, "Dynamic Behavior of Slowly-Responsive Congestion Control
Algorithms", SIGCOMM 2001, August 2001. Algorithms", SIGCOMM 2001, August 2001.
[CJ89] D. Chiu and R. Jain, "Analysis of the Increase and Decrease [CC03] Fabrizio Coccetti and Les Cottrell, TCP Stack Measurements on
Algorithms for Congestion Avoidance in Computer Networks", Computer Lightly Loaded Testbeds, 2003. URL "http://www-
Networks and ISDN Systems, Vol. 17, pp. 1-14, 1989. iepm.slac.stanford.edu/monitoring/bulk/fast/".
[CO98] J. Crowcroft and P. Oechslin, "Differentiated end-to-end [CJ89] D. Chiu and R. Jain, "Analysis of the Increase and Decrease
services using a weighted proportional fair share TCP", Computer Algorithms for Congestion Avoidance in Computer Networks", Computer
Communication Review, 28(3):53--69, 1998. Networks and ISDN Systems, Vol. 17, pp. 1-14, 1989.
[FF98] Floyd, S., and Fall, K., "Promoting the Use of End-to-End [CO98] J. Crowcroft and P. Oechslin, "Differentiated End-to-end
Congestion Control in the Internet", IEEE/ACM Transactions on Services using a Weighted Proportional Fair Share TCP", Computer
Networking, August 1999. Communication Review, 28(3):53--69, 1998.
[FRS02] Sally Floyd, Sylvia Ratnasamy, and Scott Shenker, "Modifying [D02] Tom Dunigan, Floyd's TCP slow-start and AIMD mods, URL
TCP's Congestion Control for High Speeds", May 2002. URL "http://www.csm.ornl.gov/~dunigan/net100/floyd.html".
"http://www.icir.org/floyd/notes.html".
[GRK99] Panos Gevros, Fulvio Risso and Peter Kirstein, "Analysis of a [F03] Gareth Fairey, High-Speed TCP, 2003. URL
Method for Differential TCP Service" In Proceedings of the IEEE "http://www.hep.man.ac.uk/u/garethf/hstcp/".
GLOBECOM'99, Symposium on Global Internet , December 1999, Rio de
Janeiro, Brazil.
[GV02] S. Gorinsky and H. Vin, "Extended Analysis of Binary [F92] S. Floyd and V. Jacobson, On Traffic Phase Effects in Packet-
Adjustment Algorithms", Technical Report TR2002-39, Department of Switched Gateways, Internetworking: Research and Experience, V.3
Computer Sciences, The University of Texas at Austin, August 2002. N.3, September 1992, p.115-156. URL
URL "http://www.cs.utexas.edu/users/gorinsky/pubs.html". "http://www.icir.org/floyd/papers.html".
[HSTCP] HighSpeed TCP Web Page, URL [Fl03] Sally Floyd, "Re: [Tsvwg] taking NewReno (RFC 2582) to
"http://www.icir.org/floyd/hstcp.html". Proposed Standard", Email to the tsvwg mailing list, May 14, 2003,
URLs "http://www1.ietf.org/mail-archive/working-
groups/tsvwg/current/msg04086.html" and "http://www1.ietf.org/mail-
archive/working-groups/tsvwg/current/msg04087.html".
[J02] Amit Jain and Sally Floyd, "Quick-Start for TCP and IP", [FF98] Floyd, S., and Fall, K., "Promoting the Use of End-to-End
internet draft draft-amit-quick-start-00.txt, work in progress, 2002. Congestion Control in the Internet", IEEE/ACM Transactions on
Networking, August 1999.
[K03] Tom Kelly, "Scalable TCP: Improving Performance in HighSpeed [FRS02] Sally Floyd, Sylvia Ratnasamy, and Scott Shenker, "Modifying
Wide Area Networks", February 2003. URL "http://www- TCP's Congestion Control for High Speeds", May 2002. URL
lce.eng.cam.ac.uk/~ctk21/scalable/". "http://www.icir.org/floyd/notes.html".
[M02] Matt Mathis, "Raising the Internet MTU", Web Page, URL [GRK99] Panos Gevros, Fulvio Risso and Peter Kirstein, "Analysis of
"http://www.psc.edu/~mathis/MTU/". a Method for Differential TCP Service" In Proceedings of the IEEE
GLOBECOM'99, Symposium on Global Internet , December 1999, Rio de
Janeiro, Brazil.
[RFC 1323] V. Jacobson, R. Braden, and D. Borman, TCP Extensions for [GV02] S. Gorinsky and H. Vin, "Extended Analysis of Binary
High Performance, RFC 1323, May 1992. Adjustment Algorithms", Technical Report TR2002-39, Department of
Computer Sciences, The University of Texas at Austin, August 2002.
URL "http://www.cs.utexas.edu/users/gorinsky/pubs.html".
[S02] Stanislav Shalunov, TCP Armonk, draft, 2002, URL [HSTCP] HighSpeed TCP Web Page, URL
"http://www.internet2.edu/~shalunov/tcpar/". "http://www.icir.org/floyd/hstcp.html".
[TFRC] Mark Handley, Jitendra Padhye, Sally Floyd, and Joerg Widmer, [J02] Amit Jain and Sally Floyd, "Quick-Start for TCP and IP",
TCP Friendly Rate Control (TFRC): Protocol Specification, internet internet draft draft-amit-quick-start-02.txt, work in progress,
draft draft-ietf-tsvwg-tfrc-04.txt, work in progress, 2002. 2002.
[VMSS] "Web100 at ORNL", Web Page, [K03] Tom Kelly, "Scalable TCP: Improving Performance in HighSpeed
"http://www.csm.ornl.gov/~dunigan/netperf/web100.html". Wide Area Networks", February 2003. URL "http://www-
lce.eng.cam.ac.uk/~ctk21/scalable/".
[M02] Matt Mathis, "Raising the Internet MTU", Web Page, URL
"http://www.psc.edu/~mathis/MTU/".
[Net100] The DOE/MICS Net100 project. URL
"http://www.csm.ornl.gov/~dunigan/net100/".
[NS] The NS Simulator, "http://www.isi.edu/nsnam/ns/".
[RFC 1323] V. Jacobson, R. Braden, and D. Borman, TCP Extensions for
High Performance, RFC 1323, May 1992.
[RFC3390] Allman, M., Floyd, S., and Partridge, C., "Increasing
TCP's Initial Window", RFC 3390, October 2002.
[RFC3448] Mark Handley, Jitendra Padhye, Sally Floyd, and Joerg
Widmer, TCP Friendly Rate Control (TFRC): Protocol Specification,
RFC 3448, January 2003.
[SA03] Souza, E., and Agarwal, D.A., A HighSpeed TCP Study:
Characteristics and Deployment Issues, LBNL Technical Report
LBNL-53215. URL "http://www.icir.org/floyd/hstcp.html".
[S02] Stanislav Shalunov, TCP Armonk, draft, 2002, URL
"http://www.internet2.edu/~shalunov/tcpar/".
[S03] Alex Solan, private communication, 2003.
[VMSS] "Web100 at ORNL", Web Page,
"http://www.csm.ornl.gov/~dunigan/netperf/web100.html".
[Web100] The Web100 project. URL "http://www.web100.org/".
18. Security Considerations 18. Security Considerations
This proposal makes no changes to the underlying security of TCP. This proposal makes no changes to the underlying security of TCP.
19. IANA Considerations 19. IANA Considerations
There are no IANA considerations regarding this document. There are no IANA considerations regarding this document.
A. TCP's Loss Event Rate in Steady-State 20. TCP's Loss Event Rate in Steady-State
This section gives the number of round-trip times between congestion This section gives the number of round-trip times between congestion
events for a TCP flow with D-byte packets, for D=1500, as a function events for a TCP flow with D-byte packets, for D=1500, as a function
of the connection's average throughput B in bps. To achieve this of the connection's average throughput B in bps. To achieve this
average throughput B, a TCP connection with round-trip time R in average throughput B, a TCP connection with round-trip time R in
seconds requires an average congestion window w of BR/(8D) segments. seconds requires an average congestion window w of BR/(8D) segments.
In steady-state, TCP's average congestion window w is roughly In steady-state, TCP's average congestion window w is roughly
1.2/sqrt(p) segments. This is equivalent to a lost event at most 1.2/sqrt(p) segments. This is equivalent to a lost event at most
once every 1/p packets, or at most once every 1/(pw) = w/1.5 round- once every 1/p packets, or at most once every 1/(pw) = w/1.5 round-
trip times. Substituting for w, this is a loss event at most every trip times. Substituting for w, this is a loss event at most every
(BR)/12D)round-trip times. (BR)/12D)round-trip times.
An an example, for R = 0.1 seconds and D = 1500 bytes, this gives An an example, for R = 0.1 seconds and D = 1500 bytes, this gives
B/180000 round-trip times between loss events. B/180000 round-trip times between loss events.
B. A table for a(w) and b(w). B. A table for a(w) and b(w).
This section gives a table for the increase and decrease parameters This section gives a table for the increase and decrease parameters
a(w) and b(w) for HighSpeed TCP, for the default values of Low_Window a(w) and b(w) for HighSpeed TCP, for the default values of Low_Window
= 38, High_Window = 83000, High_P = 10^-7, and High_Decrease = 0.1. = 38, High_Window = 83000, High_P = 10^-7, and High_Decrease = 0.1.
w a(w) b(w) w a(w) b(w)
---- ---- ---- ---- ---- ----
38 1 0.50 38 1 0.50
skipping to change at page 24, line 32 skipping to change at page 33, line 32
61799 65 0.12 61799 65 0.12
64851 66 0.11 64851 66 0.11
68113 67 0.11 68113 67 0.11
71617 68 0.11 71617 68 0.11
75401 69 0.10 75401 69 0.10
79517 70 0.10 79517 70 0.10
84035 71 0.10 84035 71 0.10
89053 72 0.10 89053 72 0.10
94717 73 0.09 94717 73 0.09
Table 9: Parameters for HighSpeed TCP. Table 12: Parameters for HighSpeed TCP.
This table was computed with the following Perl program: This table was computed with the following Perl program:
$top = 100000; $top = 100000;
$num = 38; $num = 38;
if ($num == 38) { if ($num == 38) {
print " w a(w) b(w)\n"; print " w a(w) b(w)\n";
print " ---- ---- ----\n"; print " ---- ---- ----\n";
print " 38 1 0.50\n"; print " 38 1 0.50\n";
$oldb = 0.50; $oldb = 0.50;
skipping to change at page 25, line 24 skipping to change at page 34, line 24
while ($num < $top) { while ($num < $top) {
$bw = (0.1 -0.5)*(log($num)-log(38))/(log(83000)-log(38))+0.5; $bw = (0.1 -0.5)*(log($num)-log(38))/(log(83000)-log(38))+0.5;
$aw = ($num**2*2.0*$bw) / ((2.0-$bw)*$num**1.2*12.8); $aw = ($num**2*2.0*$bw) / ((2.0-$bw)*$num**1.2*12.8);
if ($aw > $olda + 1) { if ($aw > $olda + 1) {
printf "%6d %5d %3.2f0, $num, $aw, $bw; printf "%6d %5d %3.2f0, $num, $aw, $bw;
$olda = $aw; $olda = $aw;
} }
$num ++; $num ++;
} }
Table 10: Perl Program for computing parameters for HighSpeed TCP. Table 13: Perl Program for computing parameters for HighSpeed TCP.
C. Exploring the time to converge to fairness. C. Exploring the time to converge to fairness.
This section gives the Perl program used to compute the congestion This section gives the Perl program used to compute the congestion
window growth during congestion avoidance. window growth during congestion avoidance.
$top = 2001; $top = 2001;
$hswin = 1; $hswin = 1;
$regwin = 1; $regwin = 1;
$rtt = 1; $rtt = 1;
skipping to change at page 26, line 30 skipping to change at page 35, line 30
} }
if ($rtt >= $lastrtt + $rttstep) { if ($rtt >= $lastrtt + $rttstep) {
printf "%5d %9d %10d0, $rtt, $hswin, $regwin; printf "%5d %9d %10d0, $rtt, $hswin, $regwin;
$lastrtt = $rtt; $lastrtt = $rtt;
} }
$hswin += $aw; $hswin += $aw;
$regwin += 1; $regwin += 1;
$rtt ++; $rtt ++;
} }
Table 11: Perl Program for computing the window in congestion Table 14: Perl Program for computing the window in congestion
avoidance. avoidance.
AUTHORS' ADDRESSES AUTHORS' ADDRESSES
Sally Floyd Sally Floyd
Phone: +1 (510) 666-2989 Phone: +1 (510) 666-2989
ICIR (ICSI Center for Internet Research) ICIR (ICSI Center for Internet Research)
Email: floyd@icir.org Email: floyd@icir.org
URL: http://www.icir.org/floyd/ URL: http://www.icir.org/floyd/
 End of changes. 167 change blocks. 
813 lines changed or deleted 1138 lines changed or added

This html diff was produced by rfcdiff 1.48. The latest version is available from http://tools.ietf.org/tools/rfcdiff/