Aditya Akella The Impact of False Sharing on Shared Congestion Management Aditya Akella with Srinivasan Seshan and Hari Balakrishnan.

Slides:



Advertisements
Similar presentations
Congestion Control and Fairness Models Nick Feamster CS 4251 Computer Networking II Spring 2008.
Advertisements

TCP--Revisited. Background How to effectively share the network? – Goal: Fairness and vague notion of equality Ideal: If N connections, each should get.
TELE202 Lecture 8 Congestion control 1 Lecturer Dr Z. Huang Overview ¥Last Lecture »X.25 »Source: chapter 10 ¥This Lecture »Congestion control »Source:
IS333, Ch. 26: TCP Victor Norman Calvin College 1.
Congestion Control Created by M Bateman, A Ruddle & C Allison As part of the TCP View project.
TCP Congestion Control Dina Katabi & Sam Madden nms.csail.mit.edu/~dina 6.033, Spring 2014.
Packet Switching Vs Circuit Switching Packet-switched and circuit-switched networks use two different technologies for sending messages and data from one.
Computer Networks: TCP Congestion Control 1 TCP Congestion Control Lecture material taken from “Computer Networks A Systems Approach”, Fourth Edition,Peterson.
Network Border Patrol: Preventing Congestion Collapse and Promoting Fairness in the Internet Celio Albuquerque, Brett J. Vickers, Tatsuya Suda 1.
Advanced Computer Networking Congestion Control for High Bandwidth-Delay Product Environments (XCP Algorithm) 1.
Chapter 6 outline r 6.1 Multimedia Networking Applications r 6.2 Streaming stored audio and video m RTSP r 6.3 Real-time, Interactive Multimedia: Internet.
XCP: Congestion Control for High Bandwidth-Delay Product Network Dina Katabi, Mark Handley and Charlie Rohrs Presented by Ao-Jan Su.
1 Estimating Shared Congestion Among Internet Paths Weidong Cui, Sridhar Machiraju Randy H. Katz, Ion Stoica Electrical Engineering and Computer Science.
The Impact of False Sharing on Shared Congestion Management Aditya Akella and Srinivasan Seshan (Computer Science Department, Carnegie Mellon University)
Metrics for Performance Evaluation Nelson Fonseca State University of Campinas.
1 Internet Networking Spring 2003 Tutorial 11 Explicit Congestion Notification (RFC 3168) Limited Transmit (RFC 3042)
Network Tomography from Multiple Senders Rob Nowak Thursday, January 15, 2004 In collaboration with Mark Coates and Michael Rabbat.
RCS: A Rate Control Scheme for Real-Time Traffic in Networks with High B X Delay and High error rates J. Tang et al, Infocom 2001 Another streaming control.
Congestion Control in Distributed Media Streaming Lin Ma Wei Tsang Ooi School of Computing National University of Singapore IEEE INFOCOM 2007.
1 Internet Networking Spring 2003 Tutorial 11 Explicit Congestion Notification (RFC 3168)
1 End-to-End Detection of Shared Bottlenecks Sridhar Machiraju and Weidong Cui Sahara Winter Retreat 2003.
On Efficient On-line Grouping of Flows with Shared Bottlenecks at Loaded Servers by O. Younis and S. Fahmy Department of Computer Sciences, Purdue University.
User-level Internet Path Diagnosis R. Mahajan, N. Spring, D. Wetherall and T. Anderson.
1 Spring Semester 2007, Dept. of Computer Science, Technion Internet Networking recitation #8 Explicit Congestion Notification (RFC 3168) Limited Transmit.
On Multi-Path Routing Aditya Akella 03/25/02. What is Multi-Path Routing?  Dynamically route traffic Multiple paths to a destination Path taken dependant.
Estimating Congestion in TCP Traffic Stephan Bohacek and Boris Rozovskii University of Southern California Objective: Develop stochastic model of TCP Necessary.
1 Reliability & Flow Control Some slides are from lectures by Nick Mckeown, Ion Stoica, Frans Kaashoek, Hari Balakrishnan, and Sam Madden Prof. Dina Katabi.
Bandwidth Measurements Jeng Lung WebTP Meeting 10/25/99.
Ningning HuCarnegie Mellon University1 A Measurement Study of Internet Bottlenecks Ningning Hu (CMU) Joint work with Li Erran Li (Bell Lab) Zhuoqing Morley.
Advanced Computer Networks: TCP Congestion Control 1 TCP Congestion Control Lecture material taken from “Computer Networks A Systems Approach”, Fourth.
L13: Sharing in network systems Dina Katabi Spring Some slides are from lectures by Nick Mckeown, Ion Stoica, Frans.
Detecting Shared Congestion of Flows Via End- to-end Measurement Dan Rubenstein Jim Kurose Don Towsley Computer Networks Research Group.
Error Checking continued. Network Layers in Action Each layer in the OSI Model will add header information that pertains to that specific protocol. On.
Multiple Sender Distributed Video Streaming Nguyen, Zakhor IEEE Transactions on Multimedia April 2004.
Ch. 28 Q and A IS 333 Spring Q1 Q: What is network latency? 1.Changes in delay and duration of the changes 2.time required to transfer data across.
NET-REPLAY: A NEW NETWORK PRIMITIVE Ashok Anand Aditya Akella University of Wisconsin, Madison.
Alok Shriram and Jasleen Kaur Presented by Moonyoung Chung Empirical Evaluation of Techniques for Measuring Available Bandwidth.
CS640: Introduction to Computer Networks Aditya Akella Lecture 20 - Queuing and Basics of QoS.
CS551: End-to-End Packet Dynamics Paxon’99 Christos Papadopoulos (
Requirements for Simulation and Modeling Tools Sally Floyd NSF Workshop August 2005.
Queueing and Active Queue Management Aditya Akella 02/26/2007.
Multiplicative Wavelet Traffic Model and pathChirp: Efficient Available Bandwidth Estimation Vinay Ribeiro.
Lecture 9 – More TCP & Congestion Control
N. Hu (CMU)L. Li (Bell labs) Z. M. Mao. (U. Michigan) P. Steenkiste (CMU) J. Wang (AT&T) Infocom 2005 Presented By Mohammad Malli PhD student seminar Planete.
1 Evaluating NGI performance Matt Mathis
CS640: Introduction to Computer Networks Aditya Akella Lecture 15 TCP – III Reliability and Implementation Issues.
Computer Networking Lecture 18 – More TCP & Congestion Control.
CS640: Introduction to Computer Networks Aditya Akella Lecture 15 TCP – III Reliability and Implementation Issues.
Analysis of RED Goal: impact of RED on loss and delay of bursty (TCP) and less bursty or smooth (UDP) traffic RED eliminates loss bias against bursty traffic.
Internet Networking recitation #11
TCP Congestion Control 컴퓨터공학과 인공지능 연구실 서 영우. TCP congestion control2 Contents 1. Introduction 2. Slow-start 3. Congestion avoidance 4. Fast retransmit.
1 Transport Layer: Basics Outline Intro to transport UDP Congestion control basics.
Spring Computer Networks1 Congestion Control Sections 6.1 – 6.4 Outline Preliminaries Queuing Discipline Reacting to Congestion Avoiding Congestion.
CS640: Introduction to Computer Networks Aditya Akella Lecture 15 TCP Congestion Control.
The Impact of False Sharing on Shared Congestion Management Srinivasa Aditya Akella Joint work with Srini Seshan and Hari Balakrishnan 28 Feb, 2001.
Tightly Coupled Congestion Control in WebRTC Michael Welzl Upperside WebRTC conference Paris
1 Three ways to (ab)use Multipath Congestion Control Costin Raiciu University Politehnica of Bucharest.
11 CS716 Advanced Computer Networks By Dr. Amir Qayyum.
1 Flow & Congestion Control Some slides are from lectures by Nick Mckeown, Ion Stoica, Frans Kaashoek, Hari Balakrishnan, and Sam Madden Prof. Dina Katabi.
Delay-based Congestion Control for Multipath TCP Yu Cao, Mingwei Xu, Xiaoming Fu Tsinghua University University of Goettingen.
Protocols and layering Network protocols and software Layered protocol suites The OSI 7 layer model Common network design issues and solutions.
Internet Networking recitation #9
Topics discussed in this section:
Lecture 19 – TCP Performance
CS640: Introduction to Computer Networks
Internet Networking recitation #10
CS4470 Computer Networking Protocols
TCP Congestion Control
An Empirical Evaluation of Wide-Area Internet Bottlenecks
Project proposal Multi-stream and multi-path audio transmission
Presentation transcript:

Aditya Akella The Impact of False Sharing on Shared Congestion Management Aditya Akella with Srinivasan Seshan and Hari Balakrishnan ICNP 2003

Aditya Akella Web Traffic Internet The Web -- Lots of concurrent flows, multiple slow starts No shared probing of network – aggressive behavior Burst losses & Inefficiency

Aditya Akella Shared Congestion Management Internet Assuming same (source, destination)  identical path, identical bottlenecks… Share congestion state, learn from each other React to losses and delays on other flows Macroflow  No more independent probes  Fewer network losses  Better ensemble behavior

Aditya Akella False Sharing Internet Flows may not share all bottlenecks  False Sharing Prioritize audio over video Flows may traverse different paths In such cases, should not share congestion state! React to slower flow – reduce speed React to faster flow – increase speed False sharing affects flow performance… Same source, destination  identical path, identical bottlenecks  QoS-enhanced Networks Multipath Routing or NATs False sharing: Two or more flows in a macroflow may not share all bottlenecks E.g., QoS-enhanced networks, Multipath routing, NATs

Aditya Akella In This Paper  How does false sharing affect flow performance?  Compromise congestion control? Or lower throughput?  How can an end-system detect false sharing?  How should end-systems respond upon detection?

Aditya Akella Outline of the Talk  Impact  Detection  Response  Summary

Aditya Akella Impact: Back-of-the-Envelop Analysis Flow 1 (RTT = R 1 ) Flow 2 (RTT = R 2 ) Src Throughput: 1 Throughput: 2  share < min( 1, 2 )  Slower sender doesn’t overwhelm its bottleneck  But faster sender could suffer badly  Extensive simulation support  Details in paper  Analysis assumes long-flows  Results similar for short-flows  False sharing does not compromise end-to-end congestion control Say Flows 1, 2 share macroflow. What is their new throughput? Dst 1 = 2 =  share

Aditya Akella Outline of the Talk  Impact  Detection  Response  Summary

Aditya Akella Detection Tests: Intuition  Flows undergoing false-sharing traverse different paths  Packets in these flows experience different base RTT, queuing and losses

Aditya Akella Detection: Out-of-Order Test Flows experience different RTT or delays on paths  Packets on different flows, sent back-to-back, will arrive out-of-order!  Out-of-Order test: Look for consistent re- ordering at receiver Increasing sequence numbers to packets in a macroflow Reordering by more than 3 packets  flag flows in macroflow If (#flags > 10), identify false-sharing Flow 1 – somewhat congested bottleneck Flow 2 – highly congested bottleneck S D Unshared Bottlenecks 2 1

Aditya Akella Detection: Delay-Correlation Test Assume sender and receiver time-stamp packets. Receiver computes  (time-stamp)  packet delay 1. Delay auto-correlation: correlation between delays of consecutive packets of a flow 2. Delay cross-correlation: correlation between delays of consecutive packets from different flows Auto-correlation > Cross-correlation  False-sharing! [Rubenstein00]

Aditya Akella Detection: Loss-Correlation Test 1. Loss auto-correlation = conditional loss probability for packet on a flow following a loss on the flow 2. Loss cross-correlation = conditional loss probability for packet on a flow following a loss on the flow Auto-correlation > Cross-correlation  False-sharing!

Aditya Akella Detection Tests: Performance  Detection accuracy  Loss test has poor accuracy  Delay test is better  Order test is the best!  Detection speed  Loss test slowest  Delay and order test fastest Unshared Bottlenecks Correct output:: “Don’t Share”’ Detection Speed Detection Accuracy SD

Aditya Akella Detection Tests: Performance  Detection speed  Order test is very fast: <5s on average  Detection accuracy  Order test: “Don’t share” > 90% of occasions  Loss, delay test: “Share” > 70% of the occasions  Summary: Out-of-order test works best  Very accurate, very fast Fully shared Bottlenecks Output from loss, delay tests: “Share” Low RTT S High RTT D Output from order test: “Don’t Share” Correct output: “Don’t Share”

Aditya Akella Outline of the Talk  Impact  Detection  Response  Summary

Aditya Akella Response to False Sharing: Design  Which is the better default? “Share” or “Don’t Share”  “Share” is a better default than “Don’t Share”  Detecting false sharing much easier  Statistically, easier to tell two things are different than to tell they are similar  Scheduler ensures packet interleaving  detection tests will work well  Out-of-order test will not work when default is “Don’t Share”

Aditya Akella Response to False Sharing  What after detection?  Stop sharing between flows!  Put flows in different macroflows  Performance of detection and response  Throughput of faster flow restored in <5s Unshared Bottlenecks SD

Aditya Akella Summary  Impact of false sharing: faster senders’ throughput could drop by a lot  Slower senders don’t overwhelm the bottlenecks on their paths  Detection: loss, delay and order based statistics can be employed  Delay statistics have better accuracy and speed than loss  Order-based tests are very fast and accurate  Response: default behavior should be to share  False-sharing is no longer a potential deployment concern…hopefully…

Aditya Akella Shared Bottlenecks: Losses and Delays Flow 1 – somewhat congested bottleneck Flow 2 – highly congested bottleneck Say flows 1, 2 share macroflow. What do their pkt losses and delay look like? (same RTT) S D Packet losses Packet delays Unshared Bottlenecks Correlation in losses/delays (or lack of it)  useful to detect false sharing!

Aditya Akella Detection: Delay-Correlation Test Assume sender and receiver time-stamp packets. Receiver computes  (time-stamp)  packet delay 1. Delay auto-correlation: correlation between delays of consecutive packets of a flow 2. Delay cross-correlation: correlation between delays of consecutive packets from different flows Auto-correlation > Cross-correlation  False-sharing! [Rubenstein00] Unsynchronized clocks? Not an issue  computation of correlations eliminates constant differences.

Aditya Akella Fully-Shared Bottlenecks  Need a new test for such situations  Idea: packets sent back-to-back will reach out-of-order  Look for consistent back-to-back arrivals at receiver: Out-of-Order test Flows face the same bottleneck Low RTT SD Fully shared Bottlenecks High RTT  Delay and loss correlation will return genuine sharing – Wrong!

Aditya Akella Detection Tests: Summary  Loss-based test  Inaccurate, very slow  Delay-based test  Quite accurate, but still somewhat slow  Out-of-order test  Very accurate and very fast  False sharing vs. genuine sharing  Markedly easier to detect false sharing  Detecting genuine sharing takes more than twice as long

Aditya Akella Delay and Loss Correlation: Practice Use 90% confidence intervals around the correlation metrics as they evolve  higher confidence Flows Share a BottleneckFlows Share no Bottlenecks Cross Correlation Auto Correlation Cross Correlation Auto Correlation Correlation measure Time in seconds  90% intervals don’t overlap anymore  quit and output result  Detecting false sharing easier (35s) than genuine sharing (100s)

Aditya Akella Detection: Out-of-Order Test Flows experience different RTT or delays on paths  Packets on different flows, sent back-to-back, will arrive out-of-order!  Out-of-Order test: Look for consistent re- ordering at receiver Increasing sequence numbers to packets in a macroflow Reordering by more than 3 packets  flag flows in macroflow If (#flags > 10), identify false-sharing Flow 1 – somewhat congested bottleneck Flow 2 – highly congested bottleneck S D Unshared Bottlenecks