idnits 2.17.1 draft-rfced-info-partridge-01.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Cannot find the required boilerplate sections (Copyright, IPR, etc.) in this document. Expected boilerplate is as follows today (2024-04-19) according to https://trustee.ietf.org/license-info : IETF Trust Legal Provisions of 28-dec-2009, Section 6.a: This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. IETF Trust Legal Provisions of 28-dec-2009, Section 6.b(i), paragraph 2: Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved. IETF Trust Legal Provisions of 28-dec-2009, Section 6.b(i), paragraph 3: This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** Missing expiration date. The document expiration date should appear on the first and last page. ** The document seems to lack a 1id_guidelines paragraph about Internet-Drafts being working documents. ** The document seems to lack a 1id_guidelines paragraph about the list of current Internet-Drafts. ** The document seems to lack a 1id_guidelines paragraph about the list of Shadow Directories. == The page length should not exceed 58 lines per page, but there was 1 longer page, the longest (page 1) being 70 lines Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an Abstract section. ** The document seems to lack a Security Considerations section. ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** The document seems to lack an Authors' Addresses Section. ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. ** There is 1 instance of too long lines in the document, the longest one being 2 characters in excess of 72. Miscellaneous warnings: ---------------------------------------------------------------------------- -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (Sep 1998) is 9348 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Missing reference section? '2' on line 179 looks like a reference -- Missing reference section? '1' on line 186 looks like a reference Summary: 11 errors (**), 0 flaws (~~), 1 warning (==), 4 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 End-To-End Research Group C. Partridge 2 INTERNET DRAFT BBN Technologies 3 Category: Informational Sep 1998 4 Expires in six months 6 ACK Spacing for High Delay-Bandwidth Paths with Insufficient Buffering 7 9 Status of this Memo 11 This document is an Internet-Draft. Internet-Drafts are working 12 documents of the Internet Engineering Task Force (IETF), its 13 areas, and its working groups. Note that other groups may also 14 distribute working documents as Internet-Drafts. 16 Internet-Drafts are draft documents valid for a maximum of six 17 months and may be updated, replaced, or obsoleted by other 18 documents at any time. It is inappropriate to use Internet- 19 Drafts as reference material or to cite them other than as 20 "work in progress." 22 To view the entire list of current Internet-Drafts, please check 23 the "1id-abstracts.txt" listing contained in the Internet-Drafts 24 Shadow Directories on ftp.is.co.za (Africa), ftp.nordu.net 25 (Northern Europe), ftp.nis.garr.it (Southern Europe), munnari.oz.au 26 (Pacific Rim), ftp.ietf.org (US East Coast), or ftp.isi.edu 27 (US West Coast). 29 An argument is made that the correct way to solve buffering shortages 30 in routers on high delay-bandwidth paths is for routers to space out 31 the TCP acks. 33 This memo presents thoughts from a discussion held at the July 1997 34 meeting of the End-To-End (E2E) Research Group. The material 35 presented is a half-baked suggestion and should not be interpreted as 36 an official recommendation of the Research Group. Comments are 37 solicited and should be addressed to the author. 39 1. Introduction 41 Suppose you want TCP implementations to be able to fill a 155 Mb/s 42 path. Further suppose that the path includes a satellite in a 43 geosynchronous orbit, so the round trip delay through the path is at 44 least 500 ms, and the delay-bandwidth product is 9.7 megabytes or 45 more. 47 If we further assume the TCP implementations support TCP Large 48 Windows and PAWS (many do), so they can manage 9.7 MB TCP window, 49 then we can be sure the TCP will eventually start sending at full 50 path rate (unless the satellite channel is very lossy). But it may 51 take a long time to get the TCP up to full speed. 53 One (of several) possible causes of the delay is a shortage of 54 buffering in routers. To understand this particular problem, 55 consider the following idealized behavior of TCP during slow start. 56 During slow start, for every segment ACKed, the sender transmits two 57 new segments. In effect, this behavior means the sender is 58 transmitting at *twice* the data rate of the segments being ACKed. 59 Keep in mind the separation between ACKs represents (in an ideal 60 world) the rate segments can flow through the bottleneck router in 61 the path. So the sender is bursting data at twice the bottleneck 62 rate, and a queue must be forming during the burst. In the simplest 63 case, the queue is entirely at the bottleneck router, and at the end 64 of the burst, the queue is storing half the data in the burst. (Why 65 half? During the burst, the sender transmitted at twice the 66 bottleneck rate. Suppose it takes one time unit to send a segment on 67 the bottlenecked link. During the burst the bottleneck will receive 68 two segments in every time unit, but only be able to transmit one 69 segment. The result is a net of one new segment queued every time 70 unit, for the life of the burst.) 72 TCP will end the slow start phase in response to the first lost 73 datagram. Assuming good quality transmission links, the first lost 74 datagram will be lost because the bottleneck queue overflowed. We 75 would like that loss to occur in the round-trip after the slow start 76 congestion window has reached the delay-bandwidth product. Now 77 consider the buffering required in the bottleneck link during the 78 next to last round trip. The sender will send an entire delay- 79 bandwidth worth of data in one-half a round-trip time (because it 80 sends at twice the channel rate). So for half the round-trip time, 81 the bottleneck router is in the mode of forwarding one segment while 82 receiving two. (For the second half of the round-trip, the router is 83 draining its queue). That means, to avoid losing any segments, the 84 router must have buffering equal to half the delay-bandwidth product, 85 or nearly 5 MB. 87 Most routers do not have anywhere near 5 MB of buffering for a single 88 link. Or, to express this problem another way, because routers do 89 not have this much buffering, the slow start stage will end 90 prematurely, when router buffering is exhausted. The consequence of 91 ending slow start prematurely is severe. At the end of slow start, 92 TCP goes into congestion avoidance, in which the window size is 93 increased much more slowly. So even though the channel is free, 94 because we did not have enough router buffering, we will transmit 95 slowly for a period of time (until the more conservative congestion 96 avoidance algorithm sends enough data to fill the channel). 98 2. What to Do? 100 So how to get around the shortage of router buffering? 102 One solution has been proposed, cascading TCPs. We would like to 103 suggest another solution, ACK spacing. Both schemes involve layer 104 violations because they require the router to examine the TCP header. 106 2.1 Cascading TCPs 108 One approach is to use cascading TCPs, in which we build a custom TCP 109 for the satellite (or bottleneck) link and insert it between the 110 sender's and receiver's TCPs, as shown below: 112 sender ---- Ground station -- satellite -- ground station -- receiver 114 +---------------+ +------------------------+ +---------------+ 115 | loop 1 | | loop 2 | | loop 3 | 116 +---------------+ +------------------------+ +---------------+ 118 This approach can work but is awkward. Among its limitations are: 119 the buffering problem remains (at points of bandwidth mismatches, 120 queues will form); the scheme violates end-to-end semantics of TCP 121 (the sender will get ACKs for data that has not and may never reach 122 the receiver); it constrains the reverse path of the TCP connection 123 to pass through points at which the multiple TCP connections are 124 spliced together (a problem if satellite links are unidirectional); 125 and it doesn't work with end-to-end encryption (i.e. if data above 126 the IP layer is encrypted). 128 2.2 ACK Spacing 130 Another approach is to find some way to spread the bursts, either by 131 having the sender spread out the segments, or having the network 132 arrange for the ACKs to arrive at the sender with a two segment 133 spacing (or larger). 135 Changing the sender is feasible, although it requires very good 136 operating system timers. But it has the disadvantage that only 137 upgraded senders get the performance improvement. 139 Finding a way for the network to space the ACKs would allow TCP 140 senders to transmit at the right rate, without modification. 141 Furthermore, it can be done by a router. The router simply has to 142 snoop the returning TCP ACKs and spread them out. (Note that if the 143 transmissions are encrypted, in many scenarios the router can still 144 figure out which segments are likely TCP ACKs and spread them out). 146 There are some difficult issues with this approach. The most notable 147 ones are: 149 1. What algorithm to use to determine the proper ACK spacing. 151 2. Related to (1), it may be necessary to known when a TCP is in 152 slow-start vs. congestion-avoidance, as the desired spacing 153 between ACKs is likely to be different in the two phases. 155 3. What to do about assymetric routes (if anything). The scheme 156 works so long as the router sees the ACKs (it does not have to see 157 the related data). However, if the ACKs do not return through the 158 ACK-spacing router, it is not possible to do ACK spacing. 160 4. How much, if at all, does ack compression between the respacing 161 point and the sender undo the effects of ack spacing? 163 5. How much per-flow (soft) state is required in the ACK spacing 164 router? 166 Despite these challenges the approach has appeal. Changing software 167 in a few routers (particularly those at likely bottleneck links) on 168 high delay-bandwidth paths could give a performance boost to lots of 169 TCP connections. 171 Security Issues 173 ACK spacing introduces no new security issues. ACK spacing does not 174 change the contents of any datagram. It simply delays some 175 datagrams in transit, just as a queue might. TCP and other higher 176 layer protocols are already required to work correctly with queueing 177 delays, and indeed, work correctly when encountering far more serious 178 transmission errors such as damage, loss, duplication and reordering 179 [2]. 181 Credit and Disclaimer 183 The particular idea of ACK spacing was developed by during the 184 meeting by Mark Handley and Van Jacobson in response to an issue 185 raised by the author, and was inspired, in part by ideas to enhance 186 wireless routers to improve TCP performance [1]. 188 Intellectual Property Issues 190 The author has learned from the IETF that parties may be attempting 191 to patent schemes similar to this one. Readers are advised to check 192 with the IETF to learn of any intellectual property rights issues. 194 References 196 1. H. Balakrishnan, V.N. Padmanabhan, S. Seshan and R.H. Katz, "A 197 Comparison of Mechanisms for Improving TCP Performance over Wireless 198 Links", Proc. ACM SIGCOMM '96, pp. 256-269. 199 2. J. Postel, ed. Transmission Control Protocol RFC-793, Internet 200 Requests for Comments, No. 793, September 1981, p. 4.