DSCP and other packet markings for WebRTC
QoS
Cisco Systems
sdhesika@cisco.com
Cisco Systems
fluffy@cisco.com
AT&T
dd5826@att.com
Cisco Systems
paulej@packetizer.com
Many networks, such as service provider and enterprise networks,
can provide treatment for individual packets based on
Differentiated Services Code Point (DSCP) values on a per-hop
basis. This document provides the recommended DSCP values for
browsers to use for various classes of traffic.
Differentiated Services Code Points (DSCP)
style packet marking can help provide QoS in some environments.
There are many use cases where such marking does not help, but it
seldom makes things worse if packets are marked appropriately. In
other words, if too many packets, say all audio or all audio and
video, are marked for a given network condition then it can prevent
desirable results. Either too much other traffic will be starved, or
there is not enough capacity for the preferentially marked packets
(i.e., audio and/or video).
This specification proposes how WebRTC applications can mark
packets. This specification does not contradict or redefine any
advice from previous IETF RFCs, but merely provides a simple set
of recommendations for implementers based on the previous RFCs
There are some environments where DSCP markings frequently help.
These include:
1. Private, wide-area networks.
2. Residential Networks. If the congested link is the broadband
uplink in a cable or DSL scenario, often residential routers/NAT
support preferential treatment based on DSCP.
3. Wireless Networks. If the congested link is a local wireless
network, marking may help.
Traditionally DSCP values have been thought of as being site
specific, with each site selecting its own code points for
controlling per-hop-behavior to influence the QoS for transport-layer
flows. However in the WebRTC use cases, the browsers need to set
them to something when there is no site specific information. In
this document, "browsers" is used synonymously with "Interactive User
Agent" as defined in the HTML specification,
. This document
describes a subset of DSCP code point values drawn from existing
RFCs and common usage for use with WebRTC applications. These
code points are solely defaults.
This specification defines some inputs that the browser in a
WebRTC application can consider to aid in determining how to set
the various packet markings and defines the mapping from
abstract QoS policies (data type, priority level) to those
packet markings.
This document exists as a complement to , which describes the interaction
between DSCP and real-time communications. It covers the
implications of using various DSCP values, particularly focusing on
Real-time Transport Protocol (RTP) streams
that are multiplexed onto a single transport-layer flow.
There are a number of guidelines specified in that should be followed when
marking traffic sent by WebRTC applications, as it is common for
multiple RTP streams to be multiplexed on the same transport-layer
flow. Generally, the RTP streams would be marked with a value
as appropriate from . A WebRTC
application might also multiplex data channel
traffic over the
same 5-tuple as RTP streams, which would also be marked as per
that table. The guidance in says that all data
channel traffic would be marked with a single value that is
typically different than the value(s) used for RTP streams
multiplexed with the data channel traffic over the same 5-tuple,
assuming RTP streams are marked with a value other than default
forwarding (DF). This is expanded upon further in the next
section.
This specification does not change or override the advice in any
other standards about setting packet markings. It simply selects
a subset of DSCP values that is relevant in the
WebRTC context. This document also specifies the inputs that
are needed by the browser to provide to the media engine.
The DSCP value set by the endpoint is not always trusted by
the network. Therefore, the DSCP value may be remarked at any
place in the network for a variety of reasons to any other DSCP
value, including default forwarding (DF) value to provide basic
best effort service. The mitigation for such action is through
an authorization mechanism. Such authorization mechanism is
outside the scope of this document. There is benefit in marking
traffic even if it only benefits the first few hops.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described
in .
The below uses the concept of a media flow, however this is
usually not equivalent to a transport-layer flow defined by a
5-tuple (source address, destination address, source port,
destination port, and protocol). Instead each media flow, such as
an RTP stream or SCTP
association carrying data channel packets
, contains all the
packets associated with an independent media entity within one
5-tuple. Specifically, a media flow is the transmitted packets
for an RTP session or an SCTP association. There may be
multiple media flows within the same 5-tuple. These media flows
might consist of different media types and have different levels
of importance to the application and, therefore, each
potentially marked using different DSCP values than for another
media flow multiplexed over the same transport-layer flow. The
following are the inputs that the browser provides to the media
engine:
Data Type: The browser provides this input as it knows if the
flow is audio, interactive video with or without audio,
non-interactive video with or without audio, or data.
Application Priority: Another input is the relative
importance of the flow within that data type. Many
applications have multiple media flows of the same data type
and often some flows are more important than others. For
example, in a video conference where there are usually audio
and video flows, the audio flow may be more important than
the video flow. JavaScript applications can tell the
browser whether a particular media flow is high, medium, low
or very low importance to the application.
defines in more
detail what an individual media flow is within the WebRTC
context.
As an example of different media flows that might be multiplexed
over the same transport-layer flow, packets related to one RTP
stream (e.g., an audio flow) carried over UDP might be one media
flow, packets related to a second RTP stream (e.g., presentation
video) carried over UDP might be a second media flow, and
finally data channel packets carried via SCTP over DTLS might be
third media flow.
Below is a table of DSCP markings for each data type of interest to
WebRTC. These DSCP values for each data type listed are a reasonable
subset of code point values taken from .
A web browser SHOULD use these values to mark the appropriate
media packets. More information on EF can be found in . More information on AF can be found in . DF is default forwarding which provides the
basic best effort service.
Data Type
Very Low
Low
Medium
High
Audio
CS1 (8)
DF (0)
EF (46)
EF (46)
Interactive Video with or without audio
CS1 (8)
DF (0)
AF42, AF43 (36, 38)
AF41, AF42 (34, 36)
Non-Interactive Video with or without audio
CS1 (8)
DF (0)
AF32, AF33 (28, 30)
AF31, AF32 (26, 28)
Data
CS1 (8)
DF (0)
AF11
AF21
The columns "very low", "low", "Medium" and "high" signify the
relative importance of the media flow within the application and
is an input that the browser receives to assist it in selecting
the DSCP value. These are referred to as application priority in
this document. Application priority does not refer to priority
in the network transport.
The above table assumes that packets marked with CS1 are treated
as "less than best effort". However, the treatment of CS1 is
implementation dependent. If an implementation treats CS1 as
other than "less than best effort", then the actual priority
(or, more precisely, the per-hop-behavior) of the packets may be
changed from what is intended. It is common for CS1 to be
treated the same as DF so anyone using CS1 cannot assume that
CS1 will be treated differently than DF. Implementers should
also note that the excess EF traffic is dropped. This could
mean that a packet marked as EF may not get through as opposed
to a packet marked with a different DSCP value.
The browser SHOULD first select the data type of the media flow.
Within the data type, the relative importance of the media flow
SHOULD be used to select the appropriate DSCP value.
The combination of data type and application priority provides
specificity and helps in selecting the right DSCP value for the
media flow. In some cases, the different drop precedence values
provides additional granularity in classifying packets within a
media flow. For example, in a video conference, the video media
flow may have medium application priority. If so, either AF42 or
AF43 may be selected. If the I-frames in the stream are more
important than the P-frames, then the I-frames can be marked
with AF42 and the P-frames marked with AF43.
All packets within a media flow SHOULD have the same application
priority. In some cases, the selected cell may have multiple
DSCP values, such as AF41 and AF42. These offer different drop
precedences. With the exception of data channel traffic, one may
select different drop precedences for the different packets in
the same media flow. Therefore, all packets in the media flow
SHOULD be marked with the same application priority, but can
have different drop precedences.
For reasons discussed in Section 6 of , if multiple media flows are
multiplexed using a reliable transport (e.g., TCP) then all of
the packets for all media flows multiplexed over that
transport-layer flow MUST be marked using the same DSCP value.
Likewise, all WebRTC data channel packets transmitted over an
SCTP association MUST be marked using the same DSCP value,
regardless of how many data channels (streams) exist or what
kind of traffic is carried over the various SCTP streams. In the
event that the browser wishes to change the DSCP value in use
for an SCTP association, it MUST reset the SCTP congestion
controller after changing values. Frequent changes in the DSCP
value used for an SCTP association are discouraged, though, as
this would defeat any attempts at effectively managing
congestion. It should also be noted that any change in DSCP
value that results in a reset of the congestion controller puts
the SCTP association back into slow start, which may have
undesirable effects on application performance.
For the data channel traffic multiplexed over an SCTP
association, it is RECOMMENDED that the DSCP value selected be
the one associated with the highest priority requested for all
data channels multiplexed over the SCTP association. Likewise,
when multiplexing multiple media flows over a TCP connection,
the DCSP value selected should be the one associated with the
highest priority requested for all multiplexed flows.
If a packet enters a QoS domain that has no support for the
above defined data types/application priority (service class),
then the network node at the edge will remark the DSCP value
based on policies. This could result in the media flow not
getting the network treatment it expects based on the original
DSCP value in the packet. Subsequently, if the packet enters a
QoS domain that supports a larger number of service classes,
there may not be sufficient information in the packet to restore
the original markings. Mechanisms for restoring such original
DSCP is outside the scope of this document.
In summary, there are no guarantees or promised level of service
with the use of DSCP. The service provided to a packet is
dependent upon the network design along the path, as well as the
congestion levels at every hop.
This specification does not add any additional security implication
other than the normal application use of DSCP. For security
implications on use of DSCP, please refer to Section 6 of RFC 4594.
Please also see as an
additional reference.
This specification does not require any actions from IANA.
This specification contains a downwards reference to . However, the parts of that RFC used by this
specification are sufficiently stable for this downward reference.
Thanks To David Black, Magnus Westerland, Paolo Severini, Jim
Hasselbrook, Joe Marcus, Erik Nordmark, and Michael Tuexen for
their help.
This document is dedicated to the memory of James Polk, a
long-time friend and colleague. James made important
contributions to this specification, including being one of its
primary authors. The IETF global community mourns his loss and
he will be missed dearly.
Note to RFC Editor: Please remove this section.
This document was originally an individual submission in RTCWeb WG.
The RTCWeb working group selected it to be become a WG document.
Later the transport ADs requested that this be moved to the TSVWG WG
as that seemed to be a better match. This document is now being
submitted as individual submission to the TSVWG with the hope that
WG will select it as a WG draft and move it forward to an RFC.
Configuration Guidelines for DiffServ Service Classes
Key words for use in RFCs to Indicate
Requirement Levels
Harvard University
1350 Mass. Ave.
Cambridge
MA 02138
- +1 617 495 3864
sob@harvard.edu
General
keyword
An Expedited Forwarding PHB (Per-Hop Behavior)
Definition of the
Differentiated Services Field (DS Field) in the IPv4 and IPv6
Headers
Cisco Systems
170 West Tasman Drive
San Jose
CA
95134-1706
USA
+1 408 525 4857
kmn@cisco.com
Torrent Networking Technologies
3000 Aerial Center
Morrisville
NC
27560
USA
+1 919 468 8466 x232
slblake@torrentnet.com
Cisco Systems
519 Lado Drive
Santa Barbara
CA
93111
USA
+1 408 526 4257
fred@cisco.com
EMC Corporation
35 Parkwood Drive
Hopkinton
MA
01748
USA
+1 508 435 1000 x76140
black_david@emc.com
Internet
internet protocol version 4
IPv6
IPv4
internet protocol version 6
type of service
Assured Forwarding PHB Group
Telia Finland
Myyrmaentie 2
Vantaa
01600
FI
jh@telia.fi
Cisco Systems
519 Lado Drive
Santa Barbara
CA
93111
US
fred@cisco.com
Lucent Technologies
300 Baker Avenue
Suite 100
Concord
MA
01742-2168
US
wweiss@lucent.com
MIT Laboratory for Computer Science
545 Technology Square
Cambridge
MA
02139
US
jtw@lcs.mit.edu