draft-bensley-tcpm-dctcp-04.txt   draft-bensley-tcpm-dctcp-05.txt 
Network Working Group S. Bensley Network Working Group S. Bensley
Internet-Draft Microsoft Internet-Draft Microsoft
Intended status: Informational L. Eggert Intended status: Informational L. Eggert
Expires: January 4, 2016 NetApp Expires: January 8, 2016 NetApp
D. Thaler D. Thaler
P. Balasubramanian P. Balasubramanian
Microsoft Microsoft
G. Judd G. Judd
Morgan Stanley Morgan Stanley
July 3, 2015 July 7, 2015
Microsoft's Datacenter TCP (DCTCP): Microsoft's Datacenter TCP (DCTCP):
TCP Congestion Control for Datacenters TCP Congestion Control for Datacenters
draft-bensley-tcpm-dctcp-04 draft-bensley-tcpm-dctcp-05
Abstract Abstract
This memo describes Datacenter TCP (DCTCP), an improvement to TCP This memo describes Datacenter TCP (DCTCP), an improvement to TCP
congestion control for datacenter traffic. DCTCP enhances Explicit congestion control for datacenter traffic. DCTCP uses improved
Congestion Notification (ECN) processing to estimate the fraction of Explicit Congestion Notification (ECN) processing to estimate the
bytes that encounter congestion, rather than simply detecting that fraction of bytes that encounter congestion, rather than simply
some congestion has occurred. DCTCP then scales the TCP congestion detecting that some congestion has occurred. DCTCP then scales the
window based on this estimate. This method achieves high burst TCP congestion window based on this estimate. This method achieves
tolerance, low latency, and high throughput with shallow-buffered high burst tolerance, low latency, and high throughput with shallow-
switches. buffered switches.
Status of This Memo Status of This Memo
This Internet-Draft is submitted in full conformance with the This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79. provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet- working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/. Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress." material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 4, 2016. This Internet-Draft will expire on January 8, 2016.
Copyright Notice Copyright Notice
Copyright (c) 2015 IETF Trust and the persons identified as the Copyright (c) 2015 IETF Trust and the persons identified as the
document authors. All rights reserved. document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of (http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License. described in the Simplified BSD License.
Table of Contents Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. DCTCP Algorithm . . . . . . . . . . . . . . . . . . . . . . . 4 3. DCTCP Algorithm . . . . . . . . . . . . . . . . . . . . . . . 4
3.1. Marking Congestion on the Switch . . . . . . . . . . . . 4 3.1. Marking Congestion on the Switch . . . . . . . . . . . . 4
3.2. Echoing Congestion Information on the Receiver . . . . . 4 3.2. Echoing Congestion Information on the Receiver . . . . . 4
3.3. Processing Congestion Indications on the Sender . . . . . 5 3.3. Processing Congestion Indications on the Sender . . . . . 5
3.4. Handling of SYN, SYN-ACK and RST Packets . . . . . . . . 7 3.4. Handling of SYN, SYN-ACK, RST Packets . . . . . . . . . . 7
4. Implementation Issues . . . . . . . . . . . . . . . . . . . . 7 4. Implementation Issues . . . . . . . . . . . . . . . . . . . . 7
5. Deployment Issues . . . . . . . . . . . . . . . . . . . . . . 7 5. Deployment Issues . . . . . . . . . . . . . . . . . . . . . . 8
6. Known Issues . . . . . . . . . . . . . . . . . . . . . . . . 8 6. Known Issues . . . . . . . . . . . . . . . . . . . . . . . . 9
7. Implementation Status . . . . . . . . . . . . . . . . . . . . 9 7. Implementation Status . . . . . . . . . . . . . . . . . . . . 10
8. Security Considerations . . . . . . . . . . . . . . . . . . . 9 8. Security Considerations . . . . . . . . . . . . . . . . . . . 10
9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 9 9. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 10
10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 9 10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 10
11. References . . . . . . . . . . . . . . . . . . . . . . . . . 9 11. References . . . . . . . . . . . . . . . . . . . . . . . . . 11
11.1. Normative References . . . . . . . . . . . . . . . . . . 9 11.1. Normative References . . . . . . . . . . . . . . . . . . 11
11.2. Informative References . . . . . . . . . . . . . . . . . 10 11.2. Informative References . . . . . . . . . . . . . . . . . 11
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 12
1. Introduction 1. Introduction
Large datacenters necessarily need a large number of network switches Large datacenters necessarily need a large number of network switches
to interconnect the servers in the datacenter. Therefore, a to interconnect the servers in the datacenter. Therefore, a
datacenter can greatly reduce its capital expenditure by leveraging datacenter can greatly reduce its capital expenditure by leveraging
low cost switches. However, low cost switches tend to have limited low cost switches. However, low cost switches tend to have limited
queue capacities and thus are more susceptible to packet loss due to queue capacities and thus are more susceptible to packet loss due to
congestion. congestion.
skipping to change at page 3, line 32 skipping to change at page 3, line 32
detect congestion. This does not meet the demands described above. detect congestion. This does not meet the demands described above.
First, the short flows will start to experience unacceptable First, the short flows will start to experience unacceptable
latencies before packet loss occurs. Second, by the time TCP latencies before packet loss occurs. Second, by the time TCP
congestion control kicks in on the sender, most of the incast burst congestion control kicks in on the sender, most of the incast burst
has already been dropped. has already been dropped.
[RFC3168] describes a mechanism for using Explicit Congestion [RFC3168] describes a mechanism for using Explicit Congestion
Notification (ECN) from the switch for early detection of congestion, Notification (ECN) from the switch for early detection of congestion,
rather than waiting for segment loss to occur. However, this method rather than waiting for segment loss to occur. However, this method
only detects the presence of congestion, not the extent. In the only detects the presence of congestion, not the extent. In the
presence of mild congestion, it reduces the TCP congestion window too presence of mild congestion, the TCP congestion window is reduced too
aggressively and unnecessarily affects the throughput of long flows. aggressively and unnecessarily affects the throughput of long flows.
Datacenter TCP (DCTCP) enhances ECN processing to estimate the Datacenter TCP (DCTCP) improvises upon traditional ECN processing by
fraction of bytes that encounter congestion, rather than simply estimating the fraction of bytes that encounter congestion, rather
detecting that some congestion has occurred. DCTCP then scales the than simply detecting that some congestion has occurred. DCTCP then
TCP congestion window based on this estimate. This method achieves scales the TCP congestion window based on this estimate. This method
high burst tolerance, low latency, and high throughput with shallow- achieves high burst tolerance, low latency, and high throughput with
buffered switches. shallow-buffered switches.
It is recommended that DCTCP be deployed in a datacenter environment
where the endpoints and the switching fabric are under a single
administrative domain. Deployment issues around coexistence of DCTCP
and conventional TCP, and lack of a negotiating mechanism between
sender and receiver, and possible mitigations are also discussed.
2. Terminology 2. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119]. document are to be interpreted as described in [RFC2119].
3. DCTCP Algorithm 3. DCTCP Algorithm
There are three components involved in the DCTCP algorithm: There are three components involved in the DCTCP algorithm:
skipping to change at page 6, line 43 skipping to change at page 6, line 47
7. Determine the end of the next observation window: 7. Determine the end of the next observation window:
DCTCP.WindowEnd = SND.NXT DCTCP.WindowEnd = SND.NXT
8. Reset the byte counters: 8. Reset the byte counters:
DCTCP.BytesSent = DCTCP.BytesMarked = 0 DCTCP.BytesSent = DCTCP.BytesMarked = 0
Rather than always halving the congestion window as described in Rather than always halving the congestion window as described in
[RFC3168], when the sender receives an indication of congestion, the [RFC3168], when the sender receives an indication of congestion
sender MUST update cwnd as follows: (ECE), the sender MUST update cwnd as follows:
cwnd = cwnd * (1 - DCTCP.Alpha / 2) cwnd = cwnd * (1 - DCTCP.Alpha / 2)
Thus, when no sent byte experienced congestion, DCTCP.Alpha equals Thus, when no sent byte experienced congestion, DCTCP.Alpha equals
zero, and cwnd is left unchanged. When all sent bytes experienced zero, and cwnd is left unchanged. When all sent bytes experienced
congestion, DCTCP.Alpha equals one, and cwnd is reduced by half. congestion, DCTCP.Alpha equals one, and cwnd is reduced by half.
Lower levels of congestion will result in correspondingly smaller Lower levels of congestion will result in correspondingly smaller
reductions to cwnd. reductions to cwnd.
Just as specified in [RFC3168], TCP should not react to congestion Just as specified in [RFC3168], TCP should not react to congestion
indications more than once every window of data. The setting of the indications more than once every window of data. The setting of the
"Congestion Window Reduced" (CWR) bit is also exactly as per "Congestion Window Reduced" (CWR) bit is also exactly as per
[RFC3168]. [RFC3168].
3.4. Handling of SYN, SYN-ACK and RST Packets 3.4. Handling of SYN, SYN-ACK, RST Packets
[RFC3168] states that "A host MUST NOT set ECT on SYN or SYN-ACK [RFC3168] requires that compliant TCP MUST NOT set ECT on SYN or SYN-
packets." [RFC5562] proposes setting ECT on SYN-ACK packets, but ACK packets. [RFC5562] proposes setting ECT on SYN-ACK packets, but
maintains the restriction of no ECT on SYN packets. Both these RFCs maintains the restriction of no ECT on SYN packets. Both these RFCs
prohibit ECT in SYN packets due to security concerns regarding prohibit ECT in SYN packets due to security concerns regarding
malicious SYN packets with ECT set. These RFCs, however, are malicious SYN packets with ECT set. These RFCs, however, are
intended for general Internet use, and do not directly apply to a intended for general Internet use, and do not directly apply to a
controlled datacenter deployment. The switching fabric can drop TCP controlled datacenter deployment. The switching fabric can drop TCP
packets that do not have the ECT set in the IP header. If SYN and packets that do not have the ECT set in the IP header. If SYN and
SYN-ACK packets for DCTCP connections are non-ECT they will be SYN-ACK packets for DCTCP connections are non-ECT they will be
dropped with high probability. For DCTCP connections SYN, SYN-ACK dropped with high probability. For DCTCP connections the sender
and RST packets are sent with ECT set. SHOULD set ECT for SYN, SYN-ACK and RST packets.
4. Implementation Issues 4. Implementation Issues
As noted in Section 3.3, the implementation must choose a suitable As noted in Section 3.3, the implementation must choose a suitable
estimation gain. [DCTCP10] provides a theoretical basis for estimation gain. [DCTCP10] provides a theoretical basis for
selecting the gain. However, it may be more practical to use selecting the gain. However, it may be more practical to use
experimentation to select a suitable gain for a particular network experimentation to select a suitable gain for a particular network
and workload. The Microsoft implementation of DCTCP in Windows and workload. The Microsoft implementation of DCTCP in Windows
Server 2012 uses a fixed estimation gain of 1/16. Server 2012 uses a fixed estimation gain of 1/16.
The implementation must also decide when to use DCTCP. Datacenter The implementation must also decide when to use DCTCP. Datacenter
servers may need to communicate with endpoints outside the servers may need to communicate with endpoints outside the
datacenter, where DCTCP is unsuitable or unsupported. Thus, a global datacenter, where DCTCP is unsuitable or unsupported. Thus, a global
configuration setting to enable DCTCP will generally not suffice. configuration setting to enable DCTCP will generally not suffice.
DCTCP may be configured based on the IP address of the remote DCTCP may be configured based on the IP address of the remote
endpoint. Microsoft Windows Server 2012 also supports automatic endpoint. Microsoft Windows Server 2012 also supports automatic
selection of DCTCP if the estimated RTT is less than 10 ms and ECN is selection of DCTCP if the estimated RTT is less than 10 msec and ECN
successfully negotiated, under the assumption that if the RTT is low, is successfully negotiated, under the assumption that if the RTT is
then the two endpoints are likely on the same datacenter network. low, then the two endpoints are likely on the same datacenter
network.
To prevent incast throughput collapse the minimum RTO (MinRTO) used
by TCP should be lowered significantly. The default value of MinRTO
in Windows is 300 msec which is much greater than the maximum
latencies inside a datacenter. Server 2012 onwards the MinRTO value
is configurable allowing values as low as 10 msec on a per subnet or
per TCP port basis or even globally. A lower MinRTO value requires
corresponding a lower delayed ACK timeout on the receiver. It is
recommended that the implementation allow configuration of lower
timeouts for DCTCP connections.
In the same vein, it is also recommended that the implementation
allow configuration of restarting the cwnd of idle DCTCP connections
as described in [RFC5681] since network conditions change rapidly in
the datacenter. The implementation can also allow configuration for
discarding the value of DCTCP.Alpha after cwnd restart and timeouts.
[RFC3168] forbids the ECN-marking of pure ACK packets because of the
inability of TCP to mitigate ACK-path congestion and protocol-wise
preferential treatment by routers. However dropping pure ACKs rather
than ECN marking them is disadvantageous in traffic scenarios typical
in the datacenter. Because of the prevalence of bursty traffic
patterns which involve transient congestion, the dropping of ACKS
causes subsequent retransmission. It is recommended that the
implementation a configuration knob that forces ECT on TCP pure ACK
packets.
5. Deployment Issues 5. Deployment Issues
DCTCP and conventional TCP congestion control does not coexist well.
In DCTCP, the marking threshold is set very low value to reduce
queueing delay, thus a relatively small amount of congestion will
exceed the marking threshold. During such periods of congestion,
conventional TCP will suffer packet losses and quickly scale back
cwnd. DCTCP, on the other hand, will use the fraction of marked
packets to scale back cwnd. Thus rate reduction in DCTCP will be
much lower than that of conventional TCP, and DCTCP traffic will
dominate conventional TCP traffic traversing the same link. Hence if
the traffic in the datacenter is a mix of conventional TCP and DCTCP,
it is recommended that DCTCP traffic be segregated from conventional
TCP traffic. [MORGANSTANLEY] describes a deployment that uses IP
DSCP bits where AQM is applied to DCTCP traffic, while TCP traffic is
managed via drop-tail queueing.
Today's commodity switches allow configuration of a different
marking/drop profile for non-TCP and non-IP packets. Non-TCP and
non-IP packets should be able to pass through the switch unless the
switch is really out of buffers. If the traffic in the datacenter
consists of such traffic (including UDP), one possible mitigation
would be to mark IP packets as ECT even when there is no transport
that is reacting to the marking.
Since DCTCP relies on congestion marking by the switch, DCTCP can Since DCTCP relies on congestion marking by the switch, DCTCP can
only be deployed in datacenters where the network infrastructure only be deployed in datacenters where the network infrastructure
supports ECN. The switches may also support configuration of the supports ECN. The switches may also support configuration of the
congestion threshold used for marking. [DCTCP10] provides a congestion threshold used for marking. The proposed parameterization
theoretical basis for selecting the congestion threshold, but as with can be configured with switches that implement RED. [DCTCP10]
estimation gain, it may be more practical to rely on experimentation provides a theoretical basis for selecting the congestion threshold,
or simply to use the default configuration of the device. but as with estimation gain, it may be more practical to rely on
experimentation or simply to use the default configuration of the
device. DCTCP will degrade to loss-based congestion control when
transiting a congested drop-tail link.
DCTCP requires changes on both the sender and the receiver, so both DCTCP requires changes on both the sender and the receiver, so both
endpoints must support DCTCP. Furthermore, DCTCP provides no endpoints must support DCTCP. Furthermore, DCTCP provides no
mechanism for negotiating its use, so both endpoints must be mechanism for negotiating its use, so both endpoints must be
configured through some out-of-band mechanism to use DCTCP. A configured through some out-of-band mechanism to use DCTCP. A
variant of DCTCP that can be deployed unilaterally and only requires variant of DCTCP that can be deployed unilaterally and only requires
standard ECN behavior has been described in [ODCTCP][BSDCAN], but standard ECN behavior has been described in [ODCTCP][BSDCAN], but
requires additional experimental evaluation. requires additional experimental evaluation.
6. Known Issues 6. Known Issues
DCTCP relies on the sender's ability to reconstruct the stream of CE DCTCP relies on the sender's ability to reconstruct the stream of CE
codepoints received by the remote endpoint. To accomplish this, codepoints received by the remote endpoint. To accomplish this,
DCTCP avoids using a single ACK packet to acknowledge segments DCTCP avoids using a single ACK packet to acknowledge segments
received both with and without the CE codepoint set. However, if an received both with and without the CE codepoint set. However, if one
ACK packet is dropped, it's possible that a subsequent ACK will or more ACK packets are dropped, it is possible that a subsequent ACK
indeed acknowledge a mix of CE and non-CE segments. This will, of will cumulatively acknowledge a mix of CE and non-CE segments. This
course, result in a less accurate congestion estimate. There are will, of course, result in a less accurate congestion estimate.
some potential mitigations: There are some potential mitigations:
o Even with a degraded congestion estimate, DCTCP may still perform o Even with a degraded congestion estimate, DCTCP may still perform
better than [RFC3168]. better than [RFC3168].
o If the estimation gain is small relative to the packet loss rate, o If the estimation gain is small relative to the packet loss rate,
the estimate may not be degraded much. the estimate may not be degraded much.
o If packet losses mostly occur under heavy congestion, most drops o If packet losses mostly occur under heavy congestion, most drops
will occur during an unbroken string of CE packets, and the will occur during an unbroken string of CE packets, and the
estimate will be unaffected. estimate will be unaffected.
However, the affect of packet drops on DCTCP under real world However, the affect of packet drops on DCTCP under real world
conditions has not been analyzed. conditions has not been analyzed.
DCTCP provides no mechanism for negotiating its use. Thus, there is DCTCP provides no mechanism for negotiating its use. Thus, there is
additional management and configuration overhead required to ensure additional management and configuration overhead required to ensure
that DCTCP is not used with non-DCTCP endpoints. The affect of using that DCTCP is not used with non-DCTCP endpoints. The affect of using
DCTCP with a standard ECN endpoint has been analyzed in DCTCP with a standard ECN endpoint has been analyzed in
[ODCTCP][BSDCAN]. Furthermore, it's possible that other [ODCTCP][BSDCAN]. Furthermore, it is possible that other
implementations may also modify [RFC3168] behavior without implementations may also modify [RFC3168] behavior without
negotiation, causing further interoperability issues. negotiation, causing further interoperability issues.
Much like standard TCP, DCTCP is biased against flows with longer Much like standard TCP, DCTCP is biased against flows with longer
RTTs. A method for improving the fairness of DCTCP has been proposed RTTs. A method for improving the fairness of DCTCP has been proposed
in [ADCTCP], but requires additional experimental evaluation. in [ADCTCP], but requires additional experimental evaluation.
7. Implementation Status 7. Implementation Status
This section documents the implementation status of the specification This section documents the implementation status of the specification
skipping to change at page 12, line 4 skipping to change at page 13, line 15
Microsoft Microsoft
Phone: +1 425 703 8835 Phone: +1 425 703 8835
Email: dthaler@microsoft.com Email: dthaler@microsoft.com
Praveen Balasubramanian Praveen Balasubramanian
Microsoft Microsoft
Phone: +1 425 538 2782 Phone: +1 425 538 2782
Email: pravb@microsoft.com Email: pravb@microsoft.com
Glenn Judd Glenn Judd
Morgan Stanley Morgan Stanley
Phone: +1 Phone: +1 973 979 6481
Email: glenn.judd@morganstanley.com Email: glenn.judd@morganstanley.com
 End of changes. 21 change blocks. 
51 lines changed or deleted 111 lines changed or added

This html diff was produced by rfcdiff 1.42. The latest version is available from http://tools.ietf.org/tools/rfcdiff/