TCPM Working Group J. Touch
Internet Draft USC/ISI
Intended status: Standards Track July 7, 2011
Expires: January 2012
Automating the Initial Window in TCP
draft-touch-tcpm-automatic-iw-01.txt
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
at any time. It is inappropriate to use Internet-Drafts as
reference material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
This Internet-Draft will expire on January 7, 2011.
Copyright Notice
Copyright (c) 2011 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://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.
Touch, (TBD) Expires January 7, 2012 [Page 1]
Internet-Draft Automating the Initial Window in TCP July 2011
Abstract
The Initial Window (IW) provides the starting point for TCP's
feedback-based congestion control algorithm. Its value has increased
over time to increase performance and to reflect increased
capability of Internet devices. This document describes a mechanism
to adjust the IW over long timescales, to make future changes more
safely deployed and to potentially avoid reexamination of this value
in the future.
Table of Contents
1. Introduction...................................................2
2. Conventions used in this document..............................3
3. Design Considerations..........................................3
4. Proposed IW Algorithm..........................................4
5. Discussion.....................................................6
6. Security Considerations........................................8
7. IANA Considerations............................................9
8. Conclusions....................................................9
9. References.....................................................9
9.1. Normative References......................................9
9.2. Informative References...................................10
10. Acknowledgments..............................................10
1. Introduction
TCP's congestion control algorithm uses an initial window value
(IW), both as a starting point for new connections and after one RTO
or more [RFC2581][RFC2861]. This value has evolved over time,
originally one maximum segment size (MSS), and increased to the
lesser of four MSS or 4,380 bytes [RFC3390][RFC5681]. For typical
Internet connections with an maximum transmission units (MTUs) of
1500 bytes, this permits three segments of 1,460 bytes each.
The IW value was originally implied in the original TCP congestion
control description, and documented as a standard in 1997
[RFC2001][Ja88]. The value was last updated in 1998 experimentally,
and moved to the standards track in 2002 [RFC2414][RFC3390]. There
have been recent proposals to update the IW based on further
increases in host and router capabilities and network capacity, some
focusing on specific values (e.g., IW=10), and others prescribing a
schedule for increases over time (e.g., IW=6 for 2011, increasing by
1-2 MSS per year).
Touch, (TBD) Expires January 7, 2012 [Page 2]
Internet-Draft Automating the Initial Window in TCP July 2011
This document proposes that TCP can objectively measure when an IW
is too large, and that such feedback should be used over long
timescales to adjust the IW automatically. The result should be
safer to deploy and might avoid the need to repeatedly revisit IW
size over time.
2. Conventions used in this document
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 RFC-2119 [RFC2119].
In this document, these words will appear with that interpretation
only when in ALL CAPS. Lower case uses of these words are not to be
interpreted as carrying RFC-2119 significance.
In this document, the characters ">>" preceding an indented line(s)
indicates a compliance requirement statement using the key words
listed above. This convention aids reviewers in quickly identifying
or finding the explicit compliance requirements of this RFC.
3. Design Considerations
TCP's IW value has existed statically for over two decades, so any
solution to adjusting the IW dynamically should have similarly
stable, non-invasive effects on the performance and complexity of
TCP. In order to be fair, the IW should be similar for most machines
on the public Internet. Finally, a desirable goal is to develop a
self-correcting algorithm, so that IW values that cause network
problems can be avoided. To that end, we propose the following list
of design goals:
o Little to no impact to TCP in the absence of loss, i.e., it
should not increase the complexity of default packet processing
in the normal case.
o Adapt to network feedback over long timescales, avoiding values
that persistently cause network problems.
We expect that, without other context, a good IW algorithm will
converge to a single value, but this is not required. An endpoint
with additional context or information, or deployed in a constrained
environment, can always use a different value. In specific,
information from previous connections, or sets of connections with a
similar path, can already be used as context for such decisions
[RFC2140].
Touch, (TBD) Expires January 7, 2012 [Page 3]
Internet-Draft Automating the Initial Window in TCP July 2011
However, if a given IW value persistently causes packet loss during
the initial burst of packets, it is clearly inappropriate and could
be inducing unnecessary loss in other competing connections. This
might happen for sites behind very slow boxes with small buffers,
which may or may not be the first hop.
4. Proposed IW Algorithm
Below is a simple description of the proposed IW algorithm. It
relies on the following parameters:
o MinIW = 3 MSS or 4,380 bytes (as per RFC3390]
o MaxIW = date.year - 2000
o MulDecr = 0.5
o AddIncr = 2 MSS
o Threshold = 0.05
We assume that the minimum IW (MinIW) should be as currently
specified [RFC3390]. The maximum IW can either be set to a fixed
value [Ch10], or set based on a schedule [Al10]. Regardless, we
propose that the value adapt over time, so have specified it in
terms of the current date. If that is not feasible or the time is
not available, a fixed value can be used. We also propose to use an
AIMD algorithm, with increase and decreases as noted.
Note that all of these parameters are up for discussion, though
should be determined by the time this document is issued as an RFC.
We do not anticipate that any of them are critical to the overall
design, especially because both current proposals are degenerate
cases of the algorithm below for given parameters (notably MulDec =
1.0 and AddIncr = 0 MSS, thus disabling the automatic part of the
algorithm).
The specific algorithm is as follows:
0. On boot:
IW = MaxIW; # assume this is in bytes, and an even number of MSS
1. Upon starting a new connection
Touch, (TBD) Expires January 7, 2012 [Page 4]
Internet-Draft Automating the Initial Window in TCP July 2011
CWND = IW;
conncount++;
IWnotchecked = 1; # true
2. If SYN-ACK includes ECN, treat as if the IW is too large
if (synackecn == 1) {
losscount++;
IWnotchecked = 0; # never check again
}
3. During a retransmission, check the seqno of the outgoing packet
(in bytes)
if (IWnotchecked && ((ISN - seqno) < IW))) {
losscount++;
IWnotchecked = 0; # never do this entire "if" again
} else {
IWnotchecked = 0; # you're beyond the IW so stop checking
}
4. Once a month or once every 1000 connections if no date is
available:
if ((monthly == TRUE) || (conncount > 1000)) {
if (losscount/conncount > threshold) {
# the number of connections with errors is too high
IW = IW * MulDecr;
} else {
IW = IW + AddIncr;
}
}
We recognize that this algorithm can yield a false positive when the
sequence number wraps around. In that case, we might be able to use
PAWS to avoid the issue, encourage the use of 64-bit sequence
numbers internal to the implementation, or ignore the issue and just
allow the false positives [RFC1323].
Standards language (as a shopping list):
MAY implement this as an alternative to RFC3390
If implemented:
MUST start IW at MaxIW - i.e., IW in the absence of other info
Touch, (TBD) Expires January 7, 2012 [Page 5]
Internet-Draft Automating the Initial Window in TCP July 2011
MUST limit MaxIW growth (Static or to year if poss)
MUST check once a month or 1,000 connections (the larger)
MUST decrease by at least 0.5x
MUST NOT increase by more than 2 MSS
SHOULD use IW that is integer multiple of 2 MSS (for ACK
compression)
MUST decrease IW if > 95% connections have errors
MAY increase IW otherwise
But MUST limit increase to 2 MSS/year (is this needed?)
SHOULD be implemented to limit performance impact
SHOULD be implemented to avoid seqno wrap issues
(anything else?)
There are some TCP connections which might not be counted at all,
such as those to/from loopback addresses, or those within the same
subnet as that of a local interface (for which congestion control is
sometimes disabled anyway). This may also include connections that
terminate before the IW is full, i.e., as a separate check at the
time of the connection closing.
The period over which the IW is updated is intended to be a long
timescale, e.g., a month or so, or 1,000 connections, whichever is
longer. An implementation might check the IW once a month, and
simply not update the IW or clear the connection counts in months
where the number of connections is too small.
5. Discussion
The following is intended as a list of notes to be discussed:
o Algorithm uses IW as an even multiple of MSS due to ACK
compression
o Impact of SEQNO wraparound vs. use of PAWS
o Algorithm now assumes bytes, not segments
Touch, (TBD) Expires January 7, 2012 [Page 6]
Internet-Draft Automating the Initial Window in TCP July 2011
o Algorithm now counts losses only during the first IW after start;
should the system ignore rechecking the burst after idle, i.e.,
do checks only once on the initial connection? To fix this, step
#2 would use "IWstart" as the front of the IW, set this to ISN at
connection start, and reset it to seqno during a slow-start
restart. This isn't a lot of code, and takes effect only during
restarts anyway - it's not in the fast path either.
o Impact of spurious retransmissions due to reordering (false
positive)
o Granularity (per-machine -same as now, per-interface? per-subnet?
vs. cost?)
o Need to keep ISN - needed for other uses (e.g., TCP-AO), and
typically kept except in Linux.
o Need for persistent state if a reboot occurs within the 1-month
window of evaluation
o Degenerate case due to failure is to act as if a fixed window,
which is what we have now
o Interaction with 2140
Basically 2140 sets CWND to something other than IW when it knows
better; this doc is for IW which is used there only for 'new' places
(or forgotten old ones).
o Explain why RWIN is not involved
Receiver-limited space
Space for reordering
NOT congestion control
Although sender window isn't useful if larger than this
CWND is a path property; RWIN is an endpoint property
o Reasons not to report-back:
- privacy concerns
- opportunity for spoofer poisoning the data (more on that in the
doc)
Touch, (TBD) Expires January 7, 2012 [Page 7]
Internet-Draft Automating the Initial Window in TCP July 2011
- using a DNS query is a bad idea
- requires every TCP stack support DNS queries
- requires a resolution step in addition to the reporting
- could cause the kernel to block on a timeout
- biased reporting
if cellphones (e.g.) never do this, we won't know about a
potentially large percent of endpoints
6. Observations
o The IW may not converge to a global value; that's OK.
o IW values can fluctuate; there should not be a significant impact
to this if that's what's seen by this algorithm.
o We do assume that losses during the IW are due to the IW being
too large; persistent errors that drop packets for other reasons
(e.g., OS bugs) can cause false positives, however this is
consistent with TCP's general assumption that loss is caused by
congestion that requires backoff. This algorithm treats the IW of
new connections as a long-timescale backoff system.
7. Security Considerations
Obvious ones - poisoning the info (fake loss, fake success), what
happens when one party disobeys, and whether anything is different
---
You can already do that within a connection too. Yes, you can
pollute aggregate info by virtue of it being aggregate. There's a
tradeoff of trust here - how much do you believe what's happening
most of the time, and how do you react to it.
If most of the connections lie about receiving data, then you see a
world where larger IW is working, and unless you detect data loss
some other way, TCP worked exactly as it should.
IMO, the good news is that:
- the IW drops if you get lots of lies about dropped packets but
then those endpoints could have just dropped the packets, and
dropping the IW is the right response anyway
Touch, (TBD) Expires January 7, 2012 [Page 8]
Internet-Draft Automating the Initial Window in TCP July 2011
- the IW increases if you get lots of lies about non-drops, but
then if you don't see anything else, you have no reason to claim
that anything is amiss anyway
So yes, there's spoofing in the aggregate. The law of large numbers
- connecting to lots of places - should help reduce that effect. But
ultimately it's not all that clear that the reactions of this sort
of poisoning aren't the right ones anyway.
In the end, it might be safer to require a high percent of
connections react badly to IW (i.e., over 95%) - that means that
a) if you did see loss, someone bad is controlling nearly all of
your connections anyway
b) if you don't see loss through TCP, and you didn't detect data
drops by other means for that many connections, you really don't
have a problem
I.e., increasing the threshold increases your ability to detect the
false IW increase case, so it's safer...
8. IANA Considerations
This document has no IANA considerations. This section should be
removed prior to publication.
9. Conclusions
<Add any conclusions>
10. References
10.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC3390] Allman, M., Floyd, S., Partridge, C., "Increasing TCP's
Initial Window", RFC 3390 (Standards Track), Oct. 2002.
[RFC5681] Allman, M., Paxson, V., Blanton, E., "TCP Congestion
Control," RFC 5681 (Standards Track), Sep. 2009.
Touch, (TBD) Expires January 7, 2012 [Page 9]
Internet-Draft Automating the Initial Window in TCP July 2011
10.2. Informative References
[Al10] Allman, M., "Initial Congestion Window Specification",
(work in progress), draft-allman-tcpm-bump-initcwnd-00,
Nov. 2010.
[Ch10] Chu, J., Dukkipati, N., Cheng, Y., Mathis, M., "Increasing
TCP's Initial Window," (work in progress), draft-ietf-
tcpm-initcwnd-01, Apr. 2011.
[Ja88] Jacobson, V., M. Karels, "Congestion Avoidance and
Control", Proc. Sigcomm 1988.
[RFC1323] Jacobson, V., Braden, R., Borman, D., "TCP Extensions for
High Performance", RFC 1323, May 1992.
[RFC2001] Stevens, W., "TCP Slow Start, Congestion Avoidance, Fast
Retransmit, and Fast Recovery Algorithms", RFC2001
(Standards Track), Jan. 1997.
[RFC2140] Touch, J., "TCP Control Block Interdependence", RFC 2140 /
STD 7(Informational), Apr. 1997.
[RFC2414] Allman, M., Floyd, S., Partridge, C., "Increasing TCP's
Initial Window", RFC 2414 (Experimental), Sept. 1998.
[RFC2581] Allman, M., Paxson, V., Stevens, W., "TCP Congestion
Control," RFC2581 (Standards Track), Apr. 1999.
[RFC2861] Handley, M., Padhye, J., Floyd, S., "TCP Congestion Window
Validation", RFC2861 (Experimental),
11. Acknowledgments
Mark Allman and Aki Nyrjinen contributed to the development of this
algorithm. Members of the TCPM mailing list also participated in
providing useful feedback.
This document was prepared using 2-Word-v2.0.template.dot.
Touch, (TBD) Expires January 7, 2012 [Page 10]
Internet-Draft Automating the Initial Window in TCP July 2011
Authors' Addresses
Joe Touch
USC/ISI
4676 Admiralty Way
Marina del Rey, CA 90292-6695 U.S.A.
Phone: +1 (310) 448-9151
Email: touch@isi.edu
Touch, (TBD) Expires January 7, 2012 [Page 11]