Network Working Group M. Allman
Request for Comments: 2581 NASA Glenn/Sterling Software
Obsoletes: 2001 V. Paxson
Category: Standards Track ACIRI / ICSI
W. Stevens
Consultant
April 1999
TCP Congestion Control
Status of this Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (1999). All Rights Reserved.
Abstract
This document defines TCP's four intertwined congestion control
algorithms: slow start, congestion avoidance, fast retransmit, and
fast recovery. In addition, the document specifies how TCP should
begin transmission after a relatively long idle period, as well as
discussing various acknowledgment generation methods.
1. IntrodUCtion
This document specifies four TCP [Pos81] congestion control
algorithms: slow start, congestion avoidance, fast retransmit and
fast recovery. These algorithms were devised in [Jac88] and [Jac90].
Their use with TCP is standardized in [Bra89].
This document is an update of [Ste97]. In addition to specifying the
congestion control algorithms, this document specifies what TCP
connections should do after a relatively long idle period, as well as
specifying and clarifying some of the issues pertaining to TCP ACK
generation.
Note that [Ste94] provides examples of these algorithms in action and
[WS95] provides an eXPlanation of the source code for the BSD
implementation of these algorithms.
This document is organized as follows. Section 2 provides various
definitions which will be used throughout the document. Section 3
provides a specification of the congestion control algorithms.
Section 4 outlines concerns related to the congestion control
algorithms and finally, section 5 outlines security considerations.
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 [Bra97].
2. Definitions
This section provides the definition of several terms that will be
used throughout the remainder of this document.
SEGMENT:
A segment is ANY TCP/IP data or acknowledgment packet (or both).
SENDER MAXIMUM SEGMENT SIZE (SMSS): The SMSS is the size of the
largest segment that the sender can transmit. This value can be
based on the maximum transmission unit of the network, the path
MTU discovery [MD90] algorithm, RMSS (see next item), or other
factors. The size does not include the TCP/IP headers and
options.
RECEIVER MAXIMUM SEGMENT SIZE (RMSS): The RMSS is the size of the
largest segment the receiver is willing to accept. This is the
value specified in the MSS option sent by the receiver during
connection startup. Or, if the MSS option is not used, 536 bytes
[Bra89]. The size does not include the TCP/IP headers and
options.
FULL-SIZED SEGMENT: A segment that contains the maximum number of
data bytes permitted (i.e., a segment containing SMSS bytes of
data).
RECEIVER WINDOW (rwnd) The most recently advertised receiver window.
CONGESTION WINDOW (cwnd): A TCP state variable that limits the
amount of data a TCP can send. At any given time, a TCP MUST NOT
send data with a sequence number higher than the sum of the
highest acknowledged sequence number and the minimum of cwnd and
rwnd.
INITIAL WINDOW (IW): The initial window is the size of the sender's
congestion window after the three-way handshake is completed.
LOSS WINDOW (LW): The loss window is the size of the congestion
window after a TCP sender detects loss using its retransmission
timer.
RESTART WINDOW (RW): The restart window is the size of the
congestion window after a TCP restarts transmission after an idle
period (if the slow start algorithm is used; see section 4.1 for
more discussion).
FLIGHT SIZE: The amount of data that has been sent but not yet
acknowledged.
3. Congestion Control Algorithms
This section defines the four congestion control algorithms: slow
start, congestion avoidance, fast retransmit and fast recovery,
developed in [Jac88] and [Jac90]. In some situations it may be
beneficial for a TCP sender to be more conservative than the
algorithms allow, however a TCP MUST NOT be more aggressive than the
following algorithms allow (that is, MUST NOT send data when the
value of cwnd computed by the following algorithms would not allow
the data to be sent).
3.1 Slow Start and Congestion Avoidance
The slow start and congestion avoidance algorithms MUST be used by a
TCP sender to control the amount of outstanding data being injected
into the network. To implement these algorithms, two variables are
added to the TCP per-connection state. The congestion window (cwnd)
is a sender-side limit on the amount of data the sender can transmit
into the network before receiving an acknowledgment (ACK), while the
receiver's advertised window (rwnd) is a receiver-side limit on the
amount of outstanding data. The minimum of cwnd and rwnd governs
data transmission.
Another state variable, the slow start threshold (ssthresh), is used
to determine whether the slow start or congestion avoidance algorithm
is used to control data transmission, as discussed below.
Beginning transmission into a network with unknown conditions
requires TCP to slowly probe the network to determine the available
capacity, in order to avoid congesting the network with an
inappropriately large burst of data. The slow start algorithm is
used for this purpose at the beginning of a transfer, or after
repairing loss detected by the retransmission timer.
IW, the initial value of cwnd, MUST be less than or equal to 2*SMSS
bytes and MUST NOT be more than 2 segments.
We note that a non-standard, experimental TCP extension allows that a
TCP MAY use a larger initial window (IW), as defined in equation 1
[AFP98]:
IW = min (4*SMSS, max (2*SMSS, 4380 bytes)) (1)
With this extension, a TCP sender MAY use a 3 or 4 segment initial
window, provided the combined size of the segments does not exceed
4380 bytes. We do NOT allow this change as part of the standard
defined by this document. However, we include discussion of (1) in
the remainder of this document as a guideline for those experimenting
with the change, rather than conforming to the present standards for
TCP congestion control.
The initial value of ssthresh MAY be arbitrarily high (for example,
some implementations use the size of the advertised window), but it
may be reduced in response to congestion. The slow start algorithm
is used when cwnd < ssthresh, while the congestion avoidance
algorithm is used when cwnd > ssthresh. When cwnd and ssthresh are
equal the sender may use either slow start or congestion avoidance.
During slow start, a TCP increments cwnd by at most SMSS bytes for
each ACK received that acknowledges new data. Slow start ends when
cwnd exceeds ssthresh (or, optionally, when it reaches it, as noted
above) or when congestion is observed.
During congestion avoidance, cwnd is incremented by 1 full-sized
segment per round-trip time (RTT). Congestion avoidance continues
until congestion is detected. One formula commonly used to update
cwnd during congestion avoidance is given in equation 2:
cwnd += SMSS*SMSS/cwnd (2)
This adjustment is executed on every incoming non-duplicate ACK.
Equation (2) provides an acceptable approximation to the underlying
principle of increasing cwnd by 1 full-sized segment per RTT. (Note
that for a connection in which the receiver acknowledges every data
segment, (2) proves slightly more aggressive than 1 segment per RTT,
and for a receiver acknowledging every-other packet, (2) is less
aggressive.)
Implementation Note: Since integer arithmetic is usually used in TCP
implementations, the formula given in equation 2 can fail to increase
cwnd when the congestion window is very large (larger than
SMSS*SMSS). If the above formula yields 0, the result SHOULD be
rounded up to 1 byte.
Implementation Note: older implementations have an additional
additive constant on the right-hand side of equation (2). This is
incorrect and can actually lead to diminished performance [PAD+98].
Another acceptable way to increase cwnd during congestion avoidance
is to count the number of bytes that have been acknowledged by ACKs
for new data. (A drawback of this implementation is that it requires
maintaining an additional state variable.) When the number of bytes
acknowledged reaches cwnd, then cwnd can be incremented by up to SMSS
bytes. Note that during congestion avoidance, cwnd MUST NOT be
increased by more than the larger of either 1 full-sized segment per
RTT, or the value computed using equation 2.
Implementation Note: some implementations maintain cwnd in units of
bytes, while others in units of full-sized segments. The latter will
find equation (2) difficult to use, and may prefer to use the
counting approach discussed in the previous paragraph.
When a TCP sender detects segment loss using the retransmission
timer, the value of ssthresh MUST be set to no more than the value
given in equation 3:
ssthresh = max (FlightSize / 2, 2*SMSS) (3)
As discussed above, FlightSize is the amount of outstanding data in
the network.
Implementation Note: an easy mistake to make is to simply use cwnd,
rather than FlightSize, which in some implementations may
incidentally increase well beyond rwnd.
Furthermore, upon a timeout cwnd MUST be set to no more than the loss
window, LW, which equals 1 full-sized segment (regardless of the
value of IW). Therefore, after retransmitting the dropped segment
the TCP sender uses the slow start algorithm to increase the window
from 1 full-sized segment to the new value of ssthresh, at which
point congestion avoidance again takes over.
3.2 Fast Retransmit/Fast Recovery
A TCP receiver SHOULD send an immediate duplicate ACK when an out-
of-order segment arrives. The purpose of this ACK is to inform the
sender that a segment was received out-of-order and which sequence
number is expected. From the sender's perspective, duplicate ACKs
can be caused by a number of network problems. First, they can be
caused by dropped segments. In this case, all segments after the
dropped segment will trigger duplicate ACKs. Second, duplicate ACKs
can be caused by the re-ordering of data segments by the network (not
a rare event along some network paths [Pax97]). Finally, duplicate
ACKs can be caused by replication of ACK or data segments by the
network. In addition, a TCP receiver SHOULD send an immediate ACK
when the incoming segment fills in all or part of a gap in the
sequence space. This will generate more timely information for a
sender recovering from a loss through a retransmission timeout, a
fast retransmit, or an experimental loss recovery algorithm, such as
NewReno [FH98].
The TCP sender SHOULD use the "fast retransmit" algorithm to detect
and repair loss, based on incoming duplicate ACKs. The fast
retransmit algorithm uses the arrival of 3 duplicate ACKs (4
identical ACKs without the arrival of any other intervening packets)
as an indication that a segment has been lost. After receiving 3
duplicate ACKs, TCP performs a retransmission of what appears to be
the missing segment, without waiting for the retransmission timer to
expire.
After the fast retransmit algorithm sends what appears to be the
missing segment, the "fast recovery" algorithm governs the
transmission of new data until a non-duplicate ACK arrives. The
reason for not performing slow start is that the receipt of the
duplicate ACKs not only indicates that a segment has been lost, but
also that segments are most likely leaving the network (although a
massive segment duplication by the network can invalidate this
conclusion). In other words, since the receiver can only generate a
duplicate ACK when a segment has arrived, that segment has left the
network and is in the receiver's buffer, so we know it is no longer
consuming network resources. Furthermore, since the ACK "clock"
[Jac88] is preserved, the TCP sender can continue to transmit new
segments (although transmission must continue using a reduced cwnd).
The fast retransmit and fast recovery algorithms are usually
implemented together as follows.
1. When the third duplicate ACK is received, set ssthresh to no more
than the value given in equation 3.
2. Retransmit the lost segment and set cwnd to ssthresh plus 3*SMSS.
This artificially "inflates" the congestion window by the number
of segments (three) that have left the network and which the
receiver has buffered.
3. For each additional duplicate ACK received, increment cwnd by
SMSS. This artificially inflates the congestion window in order
to reflect the additional segment that has left the network.
4. Transmit a segment, if allowed by the new value of cwnd and the
receiver's advertised window.
5. When the next ACK arrives that acknowledges new data, set cwnd to
ssthresh (the value set in step 1). This is termed "deflating"
the window.
This ACK should be the acknowledgment elicited by the
retransmission from step 1, one RTT after the retransmission
(though it may arrive sooner in the presence of significant out-
of-order delivery of data segments at the receiver).
Additionally, this ACK should acknowledge all the intermediate
segments sent between the lost segment and the receipt of the
third duplicate ACK, if none of these were lost.
Note: This algorithm is known to generally not recover very
efficiently from multiple losses in a single flight of packets
[FF96]. One proposed set of modifications to address this problem
can be found in [FH98].
4. Additional Considerations
4.1 Re-starting Idle Connections
A known problem with the TCP congestion control algorithms described
above is that they allow a potentially inappropriate burst of traffic
to be transmitted after TCP has been idle for a relatively long
period of time. After an idle period, TCP cannot use the ACK clock
to strobe new segments into the network, as all the ACKs have drained
from the network. Therefore, as specified above, TCP can potentially
send a cwnd-size line-rate burst into the network after an idle
period.
[Jac88] recommends that a TCP use slow start to restart transmission
after a relatively long idle period. Slow start serves to restart
the ACK clock, just as it does at the beginning of a transfer. This
mechanism has been widely deployed in the following manner. When TCP
has not received a segment for more than one retransmission timeout,
cwnd is reduced to the value of the restart window (RW) before
transmission begins.
For the purposes of this standard, we define RW = IW.
We note that the non-standard experimental extension to TCP defined
in [AFP98] defines RW = min(IW, cwnd), with the definition of IW
adjusted per equation (1) above.
Using the last time a segment was received to determine whether or
not to decrease cwnd fails to deflate cwnd in the common case of
persistent HTTP connections [HTH98]. In this case, a WWW server
receives a request before transmitting data to the WWW browser. The
reception of the request makes the test for an idle connection fail,
and allows the TCP to begin transmission with a possibly
inappropriately large cwnd.
Therefore, a TCP SHOULD set cwnd to no more than RW before beginning
transmission if the TCP has not sent data in an interval exceeding
the retransmission timeout.
4.2 Generating Acknowledgments
The delayed ACK algorithm specified in [Bra89] SHOULD be used by a
TCP receiver. When used, a TCP receiver MUST NOT excessively delay
acknowledgments. Specifically, an ACK SHOULD be generated for at
least every second full-sized segment, and MUST be generated within
500 ms of the arrival of the first unacknowledged packet.
The requirement that an ACK "SHOULD" be generated for at least every
second full-sized segment is listed in [Bra89] in one place as a
SHOULD and another as a MUST. Here we unambiguously state it is a
SHOULD. We also emphasize that this is a SHOULD, meaning that an
implementor should indeed only deviate from this requirement after
careful consideration of the implications. See the discussion of
"Stretch ACK violation" in [PAD+98] and the references therein for a
discussion of the possible performance problems with generating ACKs
less frequently than every second full-sized segment.
In some cases, the sender and receiver may not agree on what
constitutes a full-sized segment. An implementation is deemed to
comply with this requirement if it sends at least one acknowledgment
every time it receives 2*RMSS bytes of new data from the sender,
where RMSS is the Maximum Segment Size specified by the receiver to
the sender (or the default value of 536 bytes, per [Bra89], if the
receiver does not specify an MSS option during connection
establishment). The sender may be forced to use a segment size less
than RMSS due to the maximum transmission unit (MTU), the path MTU
discovery algorithm or other factors. For instance, consider the
case when the receiver announces an RMSS of X bytes but the sender
ends up using a segment size of Y bytes (Y < X) due to path MTU
discovery (or the sender's MTU size). The receiver will generate
stretch ACKs if it waits for 2*X bytes to arrive before an ACK is
sent. Clearly this will take more than 2 segments of size Y bytes.
Therefore, while a specific algorithm is not defined, it is desirable
for receivers to attempt to prevent this situation, for example by
acknowledging at least every second segment, regardless of size.
Finally, we repeat that an ACK MUST NOT be delayed for more than 500
ms waiting on a second full-sized segment to arrive.
Out-of-order data segments SHOULD be acknowledged immediately, in
order to accelerate loss recovery. To trigger the fast retransmit
algorithm, the receiver SHOULD send an immediate duplicate ACK when
it receives a data segment above a gap in the sequence space. To
provide feedback to senders recovering from losses, the receiver
SHOULD send an immediate ACK when it receives a data segment that
fills in all or part of a gap in the sequence space.
A TCP receiver MUST NOT generate more than one ACK for every incoming
segment, other than to update the offered window as the receiving
application consumes new data [page 42, Pos81][Cla82].
4.3 Loss Recovery Mechanisms
A number of loss recovery algorithms that augment fast retransmit and
fast recovery have been suggested by TCP researchers. While some of
these algorithms are based on the TCP selective acknowledgment (SACK)
option [MMFR96], such as [FF96,MM96a,MM96b], others do not require
SACKs [Hoe96,FF96,FH98]. The non-SACK algorithms use "partial
acknowledgments" (ACKs which cover new data, but not all the data
outstanding when loss was detected) to trigger retransmissions.
While this document does not standardize any of the specific
algorithms that may improve fast retransmit/fast recovery, these
enhanced algorithms are implicitly allowed, as long as they follow
the general principles of the basic four algorithms outlined above.
Therefore, when the first loss in a window of data is detected,
ssthresh MUST be set to no more than the value given by equation (3).
Second, until all lost segments in the window of data in question are
repaired, the number of segments transmitted in each RTT MUST be no
more than half the number of outstanding segments when the loss was
detected. Finally, after all loss in the given window of segments
has been successfully retransmitted, cwnd MUST be set to no more than
ssthresh and congestion avoidance MUST be used to further increase
cwnd. Loss in two successive windows of data, or the loss of a
retransmission, should be taken as two indications of congestion and,
therefore, cwnd (and ssthresh) MUST be lowered twice in this case.
The algorithms outlined in [Hoe96,FF96,MM96a,MM6b] follow the
principles of the basic four congestion control algorithms outlined
in this document.
5. Security Considerations
This document requires a TCP to diminish its sending rate in the
presence of retransmission timeouts and the arrival of duplicate
acknowledgments. An attacker can therefore impair the performance of
a TCP connection by either causing data packets or their
acknowledgments to be lost, or by forging excessive duplicate
acknowledgments. Causing two congestion control events back-to-back
will often cut ssthresh to its minimum value of 2*SMSS, causing the
connection to immediately enter the slower-performing congestion
avoidance phase.
The Internet to a considerable degree relies on the correct
implementation of these algorithms in order to preserve network
stability and avoid congestion collapse. An attacker could cause TCP
endpoints to respond more aggressively in the face of congestion by
forging excessive duplicate acknowledgments or excessive
acknowledgments for new data. Conceivably, such an attack could
drive a portion of the network into congestion collapse.
6. Changes Relative to RFC2001
This document has been extensively rewritten editorially and it is
not feasible to itemize the list of changes between the two
documents. The intention of this document is not to change any of the
recommendations given in RFC2001, but to further clarify cases that
were not discussed in detail in 2001. Specifically, this document
suggests what TCP connections should do after a relatively long idle
period, as well as specifying and clarifying some of the issues
pertaining to TCP ACK generation. Finally, the allowable upper bound
for the initial congestion window has also been raised from one to
two segments.
Acknowledgments
The four algorithms that are described were developed by Van
Jacobson.
Some of the text from this document is taken from "TCP/IP
Illustrated, Volume 1: The Protocols" by W. Richard Stevens
(Addison-Wesley, 1994) and "TCP/IP Illustrated, Volume 2: The
Implementation" by Gary R. Wright and W. Richard Stevens (Addison-
Wesley, 1995). This material is used with the permission of
Addison-Wesley.
Neal Cardwell, Sally Floyd, Craig Partridge and Joe Touch contributed
a number of helpful suggestions.
References
[AFP98] Allman, M., Floyd, S. and C. Partridge, "Increasing TCP's
Initial Window Size, RFC2414, September 1998.
[Bra89] Braden, R., "Requirements for Internet Hosts --
Communication Layers", STD 3, RFC1122, October 1989.
[Bra97] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC2119, March 1997.
[Cla82] Clark, D., "Window and Acknowledgment Strategy in TCP", RFC
813, July 1982.
[FF96] Fall, K. and S. Floyd, "Simulation-based Comparisons of
Tahoe, Reno and SACK TCP", Computer Communication Review,
July 1996. FTP://ftp.ee.lbl.gov/papers/sacks.ps.Z.
[FH98] Floyd, S. and T. Henderson, "The NewReno Modification to
TCP's Fast Recovery Algorithm", RFC2582, April 1999.
[Flo94] Floyd, S., "TCP and Successive Fast Retransmits. Technical
report", October 1994.
ftp://ftp.ee.lbl.gov/papers/fastretrans.ps.
[Hoe96] Hoe, J., "Improving the Start-up Behavior of a Congestion
Control Scheme for TCP", In ACM SIGCOMM, August 1996.
[HTH98] Hughes, A., Touch, J. and J. Heidemann, "Issues in TCP
Slow-Start Restart After Idle", Work in Progress.
[Jac88] Jacobson, V., "Congestion Avoidance and Control", Computer
Communication Review, vol. 18, no. 4, pp. 314-329, Aug.
1988. ftp://ftp.ee.lbl.gov/papers/congavoid.ps.Z.
[Jac90] Jacobson, V., "Modified TCP Congestion Avoidance Algorithm",
end2end-interest mailing list, April 30, 1990.
ftp://ftp.isi.edu/end2end/end2end-interest-1990.mail.
[MD90] Mogul, J. and S. Deering, "Path MTU Discovery", RFC1191,
November 1990.
[MM96a] Mathis, M. and J. Mahdavi, "Forward Acknowledgment: Refining
TCP Congestion Control", Proceedings of SIGCOMM'96, August,
1996, Stanford, CA. Available
fromhttp://www.psc.edu/networking/papers/papers.Html
[MM96b] Mathis, M. and J. Mahdavi, "TCP Rate-Halving with Bounding
Parameters", Technical report. Available from
http://www.psc.edu/networking/papers/FACKnotes/current.
[MMFR96] Mathis, M., Mahdavi, J., Floyd, S. and A. Romanow, "TCP
Selective Acknowledgement Options", RFC2018, October 1996.
[PAD+98] Paxson, V., Allman, M., Dawson, S., Fenner, W., Griner, J.,
Heavens, I., Lahey, K., Semke, J. and B. Volz, "Known TCP
Implementation Problems", RFC2525, March 1999.
[Pax97] Paxson, V., "End-to-End Internet Packet Dynamics",
Proceedings of SIGCOMM '97, Cannes, France, Sep. 1997.
[Pos81] Postel, J., "Transmission Control Protocol", STD 7, RFC793,
September 1981.
[Ste94] Stevens, W., "TCP/IP Illustrated, Volume 1: The Protocols",
Addison-Wesley, 1994.
[Ste97] Stevens, W., "TCP Slow Start, Congestion Avoidance, Fast
Retransmit, and Fast Recovery Algorithms", RFC2001, January
1997.
[WS95] Wright, G. and W. Stevens, "TCP/IP Illustrated, Volume 2:
The Implementation", Addison-Wesley, 1995.
Authors' Addresses
Mark Allman
NASA Glenn Research Center/Sterling Software
Lewis Field
21000 Brookpark Rd. MS 54-2
Cleveland, OH 44135
216-433-6586
EMail: mallman@grc.nasa.gov
http://roland.grc.nasa.gov/~mallman
Vern Paxson
ACIRI / ICSI
1947 Center Street
Suite 600
Berkeley, CA 94704-1198
Phone: +1 510/642-4274 x302
EMail: vern@aciri.org
W. Richard Stevens
1202 E. Paseo del Zorro
Tucson, AZ 85718
520-297-9416
EMail: rstevens@kohala.com
http://www.kohala.com/~rstevens
Full Copyright Statement
Copyright (C) The Internet Society (1999). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
