Increasing TCP’s Initial Window draft-hkchu-tcpm-initcwnd-00.txt Yuchung Cheng – [email protected] Nandita Dukkipati - [email protected] H.K. Jerry Chu - [email protected]

June, 2010

Velocity

1

Topics • • • • • •

Motivation & Justification Related Efforts Our Proposal Experimental Results Concerns Conclusion & Next Steps

2

Motivation #1 • Speed up slow start – Internet is dominated by Web traffic and short lived connections that never exit slow start – Altas Internet Observatory 2009 Annual Report (technical plenary on Thur.)

• Web objects and pages growing in size quantiles

Average 30

40

50

60

70

80

90

KB per Get

8.12

0.59

0.92

1.41

2.28

3.72

7.1

18.68

KB per Page

384

132

181

236

304

392

521

776 3

CDF of HTTP Response Sizes

4

Motivation #2 • IW=10 saves up to 4 round trips • Reverse the trend of browsers opening more and more simultaneous connections – Six per domain – IE8 is shown to open up to 180 simultaneous connections to the same server (when server advertises 30 domain names)! – Works against TCP’s congestion control mechanism – Congestion manager (CM) is difficult to implement

• Allow more fast recovery through fast retransmit 5

Justification – why is IW=10 safe? • Huge bandwidth growth since IW=4KB (1998) – Average b/w has reached 1.7Mbps world wide – Narrowband (<256Kbps) has shrunk to 5%

• Browsers open many simultaneous connections – Effectively test network with bursts much larger than IW=4KB

• TCP is already bursty – Slow start bursts pkts out at twice the bottleneck b/w

6

Related Efforts • Fast/Quick/Jump/Swifter/… Starts – Any one ready for standardization and deployment?

• Persistent HTTP – Benefit limited by connection persistency – Does not help the first data chunk, often the largest

• HTTP pipelining – Can benefit more from a larger IW – Limited deployment due to little proxy support

7

Related Efforts (cont’) • SPDY - Google’s Web experimental protocol – “An Argument For Changing TCP Slow Start” http://sites.google.com/a/chromium.org/dev/spdy/An_Ar gument_For_Changing_TCP_Slow_Start.pdf

• Congestion manager – complex to implement

• Cwnd cache – Similar to the temporal sharing of TCP states in RFC2140 but aggregated on a per /24 subnet basis

• NetDB – Global database of subnet attributes from past history 8

Our Proposal • Increase IW to 10 or higher – All experimental data shown here are from IW=10 – Ongoing experiments continue with IW=16

• Design principle - KISS – No state sharing across connections – IW a fixed value or based on data collected during 3WHS – No pacing required

• May need a non-standard response function when loss occurs during IW 9

Experiment Setup • Experiments with larger IW in several data centers over past few months • Front-end servers configuration – Linux TCP implementation, CUBIC cong. control – initcwnd option in ip route command

• Multiple connections opened by applications are served from the same data center • Results from two representative data centers for two consecutive weeks Ref: http://code.google.com/speed/articles/tcp_initcwnd_paper.pdf 10

User Network Characteristics

• Median BW – AvgDC: 1.2Mbps – SlowDC: 500Kbps • Median RTT ~ 70ms

11

Metrics of Interest and Datasets Dataset

# Subnets # Responses Vol. (TB)

AvgBaseData

1M

5.5B

39.3

AvgExpData

1M

5.5B

39.4

SlowBaseDat a

800K

1.6B

9.3

SlowExpData

800K

1.6B

9.1

• Logged HTTP transactions • Metrics – TCP Latency – Retransmission rate 12

Outline of Experiment Results • Are client receive windows large enough? • Impact of IW=10 – Overview of Web search latency – Impact of subnets of varying BW, RTT, BDP – Impact on responses of different sizes – Latency in mobile subnets – Effect on retransmission rate – Impact on applications with concurrent TCP connections 13

Client Receive Windows

receive window of first HTTP request OS

% >15KB

Average

FreeBSD

91%

58KB

iPhone

66%

87KB

Linux

6%

10KB

Mac

93%

270KB

Win 7

94%

41KB

Win Vista

94%

35KB

Win XP

88%

141KB

• Greater than 90% TCP connections have large enough receive windows to benefit from using IW=10

14

TCP Latency for Web Search

SlowDC

AvgDC

Qtls

Exp

Base

Diff %

Qtls

Exp

Base

Diff %

10

174

193

9.84%

10

204

211

3.32%

50

363

388

6.44%

50

458

474

3.38%

90

703

777

9.52%

90

1067

1194

10.64%

95

1001

1207

17.07%

95

1689

1954

13.56%

99

2937

3696

20.54%

99

5076

5986

15.20%

99.9

8463

10883

22.24%

99.9

16091

18661

13.77%

Average

514

582

11.7%

Average

751

823

8.7%

Latency measured in milliseconds 15

Latency as Functions of BW, RTT, BDP Traffic (%)

• Largest improvements (~20%) are for high RTT and BDP networks

16

Latency as Functions of BW, RTT, BDP Traffic (%)

• Slow start latency = Nslow-start * RTT + response-size/BW • Low BW subnets show significant improvements - Fewer slow start rounds, faster loss recovery 17

Latency for Varying Sizes of Responses Web Search Traffic (%)

iGoogle

Traffic (%)

• Absolute improvement increases with size • Response sizes <=3 segments perform no worse of than baseline 18

Per-subnet Latency and Mobile Networks Web Search in AvgDC Mobile subnets /24 subnet latency

Qtls

Exp

Base

Diff %

Qtls

Exp

Base

Diff %

10

301

317

5.32%

10

468

508

7.8%

50

421

450

6.89%

50

517

564

8.4%

90

943

1060

12.4%

90

1410

1699

17%

95

1433

1616

12.77%

95

2029

2414

15.9%

99

3983

4402

10.52%

99

4428

5004

11.5%

99.9

9903

11581

16.95%

99.9

9428

10639

11.4%

• Higher improvements in mobile because of larger RTTs 19

Effect on Retransmission Rate AvgDC

Exp

Base

Diff

SlowDC

Exp

Base

Diff

Web Search

1.73 [5.63] 1.55 [5.82] 0.18 [-0.2]

Web Search

3.5 [10.44] 2.98 [10.2] 0.52 [0.26]

Maps

4.17 [7.78] 3.27 [7.18] 0.9 [0.6]

Maps

5.79 [9.32] 3.94 [7.36] 1.85 [1.97]

iGoogle

1.52 [11.2] 1.17 [9.79] 0.35 [1.41]

iGoogle 2.8 [19.88] 1.88 [13.6] 0.92 [6.29]

Overall

2.29 [6.26] 1.98 [6.24] 0.31 [0.02]

Overall

4.21 [8.21] 3.54 [8.04] 0.67 [0.17]

An entry has two parts: retrx rate [% responses with >0 retrx]

• Most increase in retransmission rate from applications using multiple concurrent connections 20

Applications using Multiple Concurrent Connections Google Maps Latency AvgDC SlowDC Qtls

Exp

Base

Diff [%]

10

47

48

2.08%

50

220

225

2.22%

90

653

679

3.83%

95

1107

1143

3.15%

99

2991

3086

3.08%

99.9

7514

7792

3.57%

Qtls 10 50 90 95 96 97 99 99.9

Exp 19 170 647 1172 1401 1742 3630 10193

Base 27 176 659 1176 1396 1719 3550 9800

Diff [%] 29.6% 3.4% 1.8% 0.3% -0.4% -1.3% -2.3% -4%

• Effective IW for Maps in experiment is 80-120 segments • Latency improves on average in AvgDC and SlowDC 21

Concerns • What happen when everyone switches to IW=10? – congestion collapse unlikely since congestion backoff mechanism remains in place

• Negative impact to slow or mobile network? – Our experiments did not show much

• How does IW=10 flows affect flows with IW=3? • How does IW=10 affect non-web or long lived connections? 22

Conclusion & Next Steps • A moderate increase of IW seems to be the best “near-term” solution to relieve the slow-start logjam • Propose to TCPM for adoption as a WG item • More tests and analysis are needed! • We would like to call for volunteers to help out!

23

Increasing TCP's Initial Window Developers

Experiments with larger IW in several data centers over past ... Results from two representative data centers for .... We would like to call for volunteers to help out!

280KB Sizes 3 Downloads 271 Views

Recommend Documents

Increasing TCP's Initial Window Developers
Jul 30, 2010 - Less apples-to-apples: changes in server software and user base. Takes weeks to collect data. New methodology: Serve different IWs based ...

RFC 6928 - Increasing TCP's Initial Window - Research at Google
improves the overall performance of many web services without resulting in a congestion ..... 2 in [Duk10]), effectively limiting how much window a remote host may use. In order to ... initial window will help to stop this trend and provide the best.

An Argument for Increasing TCP's Initial Congestion Window
mining flow completion time. Our premise is that the initial congestion window should ... per domain, partly to increase parallelism and avoid head-of- line blocking of independent HTTP requests/responses, ... spread content over multiple domains so

An Argument for Increasing TCP's Initial ... Developers
page [16] showed Firefox 2.0 opened 24 connections and IE8 opened 180 ..... TCP connections, and iGoogle or Blogger photos that have relatively large ...

An Argument for Increasing TCP's Initial ... Developers
bandwidth networks also improved by a significant amount in our experiments. .... as shown in the response size distribution of Figure 1. The distribution also ...

An Argument for Increasing TCP's Initial ... Developers
cations devised their own mechanisms for faster download of. Web pages. ... (4) Allow faster recovery from losses. ..... http://code.google.com/speed/articles/.

TCP Initial Window: A Study
Nearly 90% of the Internet traffic volume is carried by TCP, the transmission control ... LAS (Least Attained Service) and PS+PS scheduling algorithms improve the ... improves the response time of small flows by avoiding slow-start penalty.

An Argument for Increasing TCP's Initial ... - Research at Google
3rd Quarter 2009. http://www.akamai.com/stateoftheinternet, 2009. [5] M. Allman, S. Floyd, and C. Partridge. Increasing TCP's. Initial Window. RFC 3390, 2002.

Characterizing the Effects of TCP's Initial Window
Email: [email protected]. Michael Welzl ... Email: [email protected] ... to a TCP sender, and timeouts significantly increase the completion time of ...

Evolution of TCP's Initial Window Size
size of 10 segments to make it a default value in the Internet hosts. They studied the performance of web traffic with larger. IW-sizes, and showed that the response times of flows are ... of flows, for deciding the best response strategy for a given

Increasing participation.pdf
critically judge the ideas as you approach a solution. 9) Clustering is an alternative to brainstorming. To do. this, begin with a word, name, or concept written in ...

Initial pages
May 30, 2014 - Narayana Murthy of Infosys. Unit II Wit and Humour. Wit and humour are indispensible though some consider these two are distinct entities. Wit is the ability to say or write things that are both clever and amusing. Sometimes it is foun

Contract Advisory Systems Developers and Systems Developers ...
Conducts and/or participates in Operability and System Integration testing of ... Contract Advisory Systems Developers and Systems Developers 2015.pdf.

Exterminating inefficiencies, increasing ROI
The ability to report and optimize based on the offline impact of paid search is important for ... specific keyword by search engine in almost real time. The ability to ...

One Window Service.pdf
Sign in. Loading… Whoops! There was a problem loading more pages. Whoops! There was a problem previewing this document. Retrying... Download. Connect ...

overton window pdf
File: Overton window pdf. Download now. Click here if your download doesn't start automatically. Page 1. overton window pdf. overton window pdf. Open. Extract.

Window Notes Freebie.pdf
Page 1 of 6. Window Notes. on Spring Weather. Facts about spring weather: What does the idiom: “In like a lion, out. like a lamb” mean to you? Hurtful Spring Weather: Helpful Spring Weather: Name__________. While reading the article on Spring wea

X Window System Protocol - XFree86
standards, such as the Inter-Client Communication Conventions Manual and the X Logical Font. Description .... the depth, in which case the least significant bits are used to hold the pixmap data, and the values of the unused ...... True and the bytes

ipad initial setup.pdf
Loading… Whoops! There was a problem loading more pages. Retrying... Whoops! There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. ipad initial setup.pdf. ipad

TCPS Controller Design Using Fuzzy Logic Controller for Power ...
For economic and ecological reasons, the building of new transmission lines and expansion of existing transmission systems are becoming more and more difficult. In this new situation, it is necessary to utilize the existing power transmission system

Increasing cost pressure
Rail rings an alarm. SMRT's rail operations recorded its first ever loss in 4QFY15, due mainly to increased R&M cost. Management highlighted the need to ..... 21. 21. Total Non-current Assets. 2,132. 2,143. 1,889. 1,898. Short-term Debt. 9. 9. 9. 9.

lockaire window brochure.pdf
Retrying... lockaire window brochure.pdf. lockaire window brochure.pdf. Open. Extract. Open with. Sign In. Main menu. Displaying lockaire window brochure.pdf.

window comparator pdf
Page 1. Whoops! There was a problem loading more pages. window comparator pdf. window comparator pdf. Open. Extract. Open with. Sign In. Main menu.