HyStart++: Modified Slow Start for TCP
draft-ietf-tcpm-hystartplusplus-00

Abstract

This doument describes HyStart++, a simple modification to the slow start phase of TCP congestion control algorithms. Traditional slow start can cause overshotting of the ideal send rate and cause large packet loss within a round-trip time which results in poor performance. HyStart++ combines the use of one variant of HyStart and Limited Slow Start (LSS) to prevent overshooting of the ideal sending rate, while also mitigating poor performance which can result from false positives when HyStart is used alone.

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). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

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."

This Internet-Draft will expire on January 14, 2021.

Copyright Notice

Copyright (c) 2020 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 (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.

Table of Contents

• 1. Introduction
• 2. Terminology
• 3. Definitions
• 4. HyStart++ Algorithm
• 4.1. Use of HyStart Delay Increase and Limited Slow Start
• 4.2. Algorithm Details
• 4.3. Tuning constants
• 5. Deployments and Performance Evaluations
• 6. Security Considerations
• 7. IANA Considerations
• 8. Acknowledgements
• 9. References
• 9.1. Normative References
• 9.2. Informative References
• Authors' Addresses

1. Introduction

[RFC5681] describes the slow start congestion control algorithm for TCP. The slow start algorithm is used when the congestion window (cwnd) is less than the slow start threshold (ssthresh). During slow start, in absence of packet loss signals, TCP sender increases cwnd exponentially to probe the network capacity. Such a fast growth can lead to overshooting the ideal sending rate and cause significant packet loss. This is counter-productive for the TCP flow itself, and also impacts the rest of the traffic sharing the bottleneck link. TCP has several mechanisms for loss recovery, but they are only effective for moderate loss. When these techniques are unable to recover lost packets, a last-resort retransmission timeout (RTO) is used to trigger packet recovery. In most operating systems, the minimum RTO is set to a large value (200 msec or 300 msec) to prevent spurious timeouts. This results in a long idle time which drastically impairs flow completion times.

HyStart++ adds delay increase as a signal to exit slow start before any packet loss occurs. This is one of two algorithms specified in [HyStart]. After the HyStart delay algorithm finds an exit point, LSS is used in conjunction with congestion avoidance for further congestion window increases until the first packet loss is detected. HyStart++ reduces packet loss and retransmissions, and improves goodput in lab measurements as well as real world deployments.

2. Terminology

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 [RFC2119].

3. Definitions

We repeat here some definition from [RFC5681] to aid the reader.

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 [RFC1191, RFC4821] 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, it is 536 bytes [RFC1122]. The size does not include the TCP/IP headers and options.

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.

4. HyStart++ Algorithm

4.1. Use of HyStart Delay Increase and Limited Slow Start

[HyStart] specifies two algorithms (a “Delay Increase” algorithm and an “Inter-Packet Arrival” algorithm) to be run in parallel to detect that the sending rate has reached capacity. In practice, the Inter-Packet Arrival algorithm does not perform well and is not able to detect congestion early, primarily due to ACK compression. The idea of the Delay Increase algorithm is to look for RTT spikes, which suggest that the bottleneck buffer is filling up.

After the HyStart “Delay Increase” algorithm triggers an exit from slow start, LSS (described in [RFC3742]) is used to increase Cwnd until congestion is observed. LSS is used because the HyStart exit is often premature as a result of RTT fluctuations or transient queue buildup. LSS grows the cwnd fast but much slower than traditional slow start. LSS helps avoid massive packet losses and subsequent time spent in loss recovery or retransmission timeout.

4.2. Algorithm Details

We assume that Appropriate Byte Counting (as described in [RFC3465]) is in use and L is the cwnd increase limit. The choice of value of L is up to the implementation.

A round is chosen to be approximately the Round-Trip Time (RTT). Round can be approximated using sequence numbers as follows:

• Define windowEnd as a sequence number initialize to SND.UNA
• When windowEnd is ACKed, the current round ends and windowEnd is set to SND.NXT

At the start of each round during slow start:

• lastRoundMinRTT = currentRoundMinRTT
• currentRoundMinRTT = infinity
• rttSampleCount = 0

For each arriving ACK in slow start, where N is the number of previously unacknowledged bytes acknowledged in the arriving ACK and w:

• Update the cwnd
  • cwnd = cwnd + min (N, L * SMSS)

• Keep track of minimum observed RTT
  • currentRoundMinRTT = min(currentRoundMinRTT, currRTT)
  • where currRTT is the measured RTT based on the incoming ACK
  • rttSampleCount += 1

• For rounds where cwnd is at or higher than LOW_CWND and N_RTT_SAMPLE RTT samples have been obtained, check if delay increase triggers slow start exit
  • if (cwnd >= (LOW_CWND * SMSS) AND rttSampleCount >= N_RTT_SAMPLE)
    • RttThresh = clamp(MIN_RTT_THRESH, lastRoundMinRTT / 8, MAX_RTT_THRESH)
    • if (currentRoundMinRTT >= (lastRoundMinRTT + RttThresh))
      • ssthresh = cwnd
      • exit slow start and enter LSS

For each arriving ACK in LSS, where N is the number of previously unacknowledged bytes acknowledged in the arriving ACK:

• K = cwnd / (LSS_DIVISOR * ssthresh)
• cwnd = max(cwnd + (min (N, L * SMSS) / K), CA_cwnd())

CA_cwnd() denotes the cwnd that a congestion control algorithm would have increased to if congestion avoidance started instead of LSS. LSS grows cwnd very fast but for long-lived flows in high BDP networks, the congestion avoidance algorithm could increase cwnd much faster. For example, CUBIC congestion avoidance [RFC8312] in convex region can ramp up cwnd rapidly. Taking the max can help improve performance when exiting slow start prematurely.

HyStart++ ends when congestion is observed.

4.3. Tuning constants

It is RECOMMENDED that a HyStart++ implementation use the following constants:

• LOW_CWND = 16
• MIN_RTT_THRESH = 4 msec
• MAX_RTT_THRESH = 16 msec
• LSS_DIVISOR = 0.25
• N_RTT_SAMPLE = 8

These constants have been determined with lab measurements and real world deployments. An implementation MAY tune them for different network characteristics.

Using smaller values of LOW_CWND will cause the algorithm to kick in before the last round RTT can be measured, particularly if the implementation uses an initial cwnd of 10 MSS. Higher values will delay the detection of delay increase and reduce the ability of HyStart++ to prevent overshoot problems.

The delay increase sensitivity is determined by MIN_RTT_THRESH and MAX_RTT_THRESH. Smaller values of MIN_RTT_THRESH may cause spurious exits from slow start. Larger values of MAX_RTT_THRESH may result in slow start not exiting until loss is encountered for connections on large RTT paths.

A TCP implementation is required to take at least one RTT sample each round. Using lower values of N_RTT_SAMPLE will lower the accuracy of the measured RTT for the round; higher values will improve accuracy at the cost of more processing.

The maximum value of LSS_DIVISOR SHOULD NOT exceed 0.5, which is the value recommended in [RFC3742]. Otherwise the cwnd growth could again become too aggressive and cause ideal send rate overshoot. Smaller values will cause the algorithm to be less aggressive and may leave some cwnd growth on the table.

An implementation SHOULD use HyStart++ only for the initial slow start and fall back to using traditional slow start for the remainder of the connection lifetime. This is acceptable because subsequent slow starts will use the discovered ssthresh value to exit slow start. An implementation MAY use HyStart++ to grow the restart window ([RFC5681]) after a long idle period.

5. Deployments and Performance Evaluations

As of the time of writing, HyStart++ has been default enabled for all TCP connections in Windows for two years. The original Hystart has been default-enabled for all TCP connections in Linux TCP for a decade.

In lab measurements with Windows TCP, HyStart++ shows both goodput improvements as well as reductions in packet loss and retransmissions. For example across a variety of tests on a 100 Mbps link with a bottleneck buffer size of bandwidth-delay product, HyStart++ reduces bytes retransmitted by 50% and retransmission timeouts by 36%.

In an A/B test across a large Windows device population, out of 52 billion TCP connections, 0.7% of connections move from 1 RTO to 0 RTOs and another 0.7% connections move from 2 RTOs to 1 RTO with HyStart++. This test did not focus on send heavy connections and the impact on send heavy connections is likely much higher. We plan to conduct more such production experiments to gather more data in the future.

6. Security Considerations

HyStart++ enhances slow start and inherits the general security considerations discussed in [RFC5681].

7. IANA Considerations

This document has no actions for IANA.

8. Acknowledgements

Neal Cardwell suggested the idea of using the maximum of cwnd value computed by LSS and congestion avoidance after exiting slow start.

9. References

9.1. Normative References

9.2. Informative References

Authors' Addresses