LISA: A Linked Slow-Start Algorithm for MPTCP draft-barik-mptcp-lisa-01

Similar documents
BLEST: Blocking Estimation-based MPTCP Scheduler for Heterogeneous Networks. Simone Ferlin, Ozgu Alay, Olivier Mehani and Roksana Boreli

TCP with dynamic FEC For High Delay and Lossy Networks. Simone Ferlin and Ozgu Alay Simula Research Laboratory, Norway

Implementation and Evaluation of Coupled Congestion Control for Multipath TCP

Probe or Wait : Handling tail losses using Multipath TCP

Multipath QUIC: Design and Evaluation

Multipath QUIC: Design and Evaluation

Implementation Experiments on HighSpeed and Parallel TCP

One More Bit Is Enough

Tuning the Aggressive Slow-Start Behavior of MPTCP for Short Flows

Why Your Application only Uses 10Mbps Even the Link is 1Gbps?

Congestion Control of MPTCP: Performance Issues and a Possible Solution

TCP Modifications for Congestion Exposure

Model-based Measurements Of Network Loss

Progress on Practical Shared Bottleneck Detection for Coupled Congestion Control

WB-RTO: A Window-Based Retransmission Timeout. Ioannis Psaras, Vassilis Tsaoussidis Demokritos University of Thrace, Xanthi, Greece

Intended status: Standards Track. M. Welzl. University of Oslo. June 27, 2016

MPTCP: Design and Deployment. Day 11

STMS: Improving MPTCP Throughput Under Heterogenous Networks

Reasons not to Parallelize TCP Connections for Fast Long-Distance Networks

SharkFest 17 Europe. My TCP ain t your TCP. Simon Lindermann. Stack behavior back then and today. Miele & Cie KG.

TCP SIAD: Congestion Control supporting Low Latency and High Speed

Studying Congestion Control with Explicit Router Feedback Using Hardware-based Network Emulator

Implementing and Experimenting with XCP

Exercises TCP/IP Networking With Solutions

TCP Friendly Rate Control (TFRC): Protocol Specification RFC3448bis

Performance-oriented Congestion Control

8. TCP Congestion Control

CS268: Beyond TCP Congestion Control

Performance-oriented Congestion Control

Combating Packet Reordering in Vertical Handoff using Cross-layer Notifications to TCP

OpenTCP: Combining Congestion Controls of Parallel TCP Connections

RMCAT - Shared Bottleneck Detection and Coupled Congestion Control vs. QoS for WebRTC Michael Welzl

Project Details. Jiasi Chen CS 179i: Project in Computer Science (Networks) Lectures: Monday 3:10-4pm in Spieth 1307

Master s Thesis. TCP Congestion Control Mechanisms for Achieving Predictable Throughput

Congestion Control Without a Startup Phase

Coupled Conges,on Control for RTP Media. Safiqul Islam, Michael Welzl, Stein Gjessing and Naeem Khademi Department of Informa,cs University of Oslo

Congestion Control for High Bandwidth-Delay Product Networks

TCP modifications for Congestion Exposure

Flow and Congestion Control Marcos Vieira

An Evaluation of Adaptive Multimedia Communication from a QoS Perspective

CSCI-1680 Transport Layer III Congestion Control Strikes Back Rodrigo Fonseca

Transport Over IP. CSCI 690 Michael Hutt New York Institute of Technology

Recap. TCP connection setup/teardown Sliding window, flow control Retransmission timeouts Fairness, max-min fairness AIMD achieves max-min fairness

CS519: Computer Networks. Lecture 5, Part 4: Mar 29, 2004 Transport: TCP congestion control

Congestion Control For Coded TCP. Doug Leith

file:///c:/users/hpguo/dropbox/website/teaching/fall 2017/CS4470/H...

XCP: explicit Control Protocol

Making TCP more Robust against Packet Reordering

How$speedy$is$SPDY? Xiao%(Sophia)%Wang,$Aruna$Balasubramanian,$Arvind$ Krishnamurthy,$David$Wetherall$ University*of*Washington*

HighSpeed TCP for Large Congestion Windows draft-floyd-tcp-highspeed-00.txt

Internet Engineering Task Force. Intended status: Informational Expires: August 20, 2012 Technical University Berlin February 17, 2012

Congestion control mechanism of TCP for achieving predictable throughput

Design and Performance Evaluation of High Efficient TCP for HBDP Networks

Revisiting Congestion Control for Multipath TCP with Shared Bottleneck Detection

QUIC evaluation. HTTP Workshop. 28 July 2015 Münster - Germany. G. Carlucci, L. De Cicco, S. Mascolo. Politecnico di Bari, Italy

BBR Congestion Control: IETF 100 Update: BBR in shallow buffers

Communication Networks

CS644 Advanced Networks

Characterizing Slow Start in High Speed Networks

Experimental Evaluation of Multipath TCP Schedulers

CUBIC. Qian HE (Steve) CS 577 Prof. Bob Kinicki

Computer Networking Introduction

Testing the Channel Aggregation Capability of the MPT Multipath Communication Library

Performance Comparison of TFRC and TCP

6.033 Computer System Engineering

MPTCP Application Considerations. draft-scharf-mptcp-api-00.txt. Michael Scharf Alan Ford

Utilizing Datacenter Networks: Centralized or Distributed Solutions?

Outline Computer Networking. TCP slow start. TCP modeling. TCP details AIMD. Congestion Avoidance. Lecture 18 TCP Performance Peter Steenkiste

Report on Transport Protocols over Mismatched-rate Layer-1 Circuits with 802.3x Flow Control

SharkFest'17 US. Understanding Throughput & TCP Windows. A Walk-Through of the Factors that can limit TCP Throughput Performance

Measuring end-to-end bandwidth with Iperf using Web100

An Issue in NewReno After Fast Recovery. Yoshifumi Nishida

Chapter 3 Transport Layer

Transmission Control Protocol. ITS 413 Internet Technologies and Applications

Marooned Magic Numbers An Adaptive Congestion Control Architecture

Lecture 15: Transport Layer Congestion Control

TCP Congestion Control

TCP Congestion Control

CTCP (Circuit TCP) v1.0

Improving the Robustness of TCP to Non-Congestion Events

Congestion Avoidance and Control. Rohan Tabish and Zane Ma

TCP modifications for Congestion Exposure

Fall 2012: FCM 708 Bridge Foundation I

Hybrid Control and Switched Systems. Lecture #17 Hybrid Systems Modeling of Communication Networks

Designing a Resource Pooling Transport Protocol

Brief Overview and Background

TCP Congestion Control 65KB W

Measuring end-to-end bandwidth with Iperf using Web100

TCP Congestion Control

CONCURRENT MULTIPATH TRANSFER USING TRANSPORT LAYER MULTIHOMING: PERFORMANCE UNDER VARYING BANDWIDTH PROPORTIONS

A Hybrid Systems Modeling Framework for Fast and Accurate Simulation of Data Communication Networks. Motivation

A proposal for MPTCP Robust session Establishment (MPTCP RobE) enable full multipath capability for MPTCP Markus Amend, Eckard Bogenfeld, Andreas

INTERNATIONAL JOURNAL OF ADVANCED RESEARCH IN ENGINEERING AND TECHNOLOGY (IJARET)

Congestion Control In The Internet Part 2: How it is implemented in TCP. JY Le Boudec 2014

Topics. TCP sliding window protocol TCP PUSH flag TCP slow start Bulk data throughput

TCP OVER AD HOC NETWORK

Experimental Evaluation of Multipath TCP Schedulers

Implementing TCP SACK Conservative Loss Recovery Algorithm within a NDN Consumer

0-RTT TCP Convert Protocol

Congestion Control In The Internet Part 2: How it is implemented in TCP. JY Le Boudec 2014

Transcription:

LISA: A Linked Slow-Start Algorithm for draft-barik-mptcp-lisa-1 Runa Barik (UiO), Simone Ferlin (SRL), Michael Welzl (UiO) Multipath TCP @96th IETF Meeting Berlin, Germany th July 16 IETF96 LISA: A Linked Slow-Start Algorithm for 1 / 16

What is new? A paper 1 published. A patch file on http://heim.ifi.uio.no/runabk/lisa/ Updated by draft-barik-mptcp-lisa-1 Presented in IETF94, Yokahama, Japan. 1 Runa Barik, Michael Welzl, Simone Ferlin and Ozgu Alay, LISA: A Linked Slow-Start Algorithm for, in IEEE ICC 16 IETF96 LISA: A Linked Slow-Start Algorithm for 2 / 16

is very simple; it is just an update to the very first slow-start of new subflows. When a new subflow joins, we find one available subflow that could give a part of its cwnd to the new subflow. If there is no subflow, assign the default values based on RFC 339 and RFC 6928. The goal is to reduce temporary aggressiveness, and losses at the end of slow-start. IETF96 LISA: A Linked Slow-Start Algorithm for 3 / 16

Outline Feedback from IETF94: Slowing down slow start will reduce retransmissions, does not work in large delay-bw environments. See graphs relating to BDP. Current parameters are unrealistic. What is now? Performance of LISA in Shared and Non-Shared Bottleneck, varying: Number of subflows RTT Bandwidth LISA behavior for a large transfer IETF96 LISA: A Linked Slow-Start Algorithm for 4 / 16

Topology Server Eth Eth1 Eth Bottleneck Link Client (a) Shared bottleneck Bottleneck Link Server Bottleneck Link (b) Non-Shared bottleneck Client IETF96 LISA: A Linked Slow-Start Algorithm for 5 / 16

Shared Bottleneck: Number of subflows 1 8 Base RTT=ms BW=5Mbps 7 8 9 (c) Mean-completion time (2 subflows) 7 7 8 9 (e) Retransmissions 2 18 1 8 BW=5Mbps Base RTT=ms 8 1 (d) Mean-completion time (4 subflows) 8 8 1 (f) Retransmissions IETF96 LISA: A Linked Slow-Start Algorithm for 6 / 16

Shared Bottleneck: RTT Base RTT=1ms BW=5Mbps 7 9 1 17 19 (a) Mean-completion time 1 8 7 9 1 17 19 (b) Retransmissions sum of cwnd 1 1 8 Transfer-size=KB.2.4.6.8 1 1.2 1.4.2.4.6.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 Time (in sec) Time (in sec) IETF96 LISA: A Linked Slow-Start Algorithm for 7 / 16 sum of cwnd 18 1 1 1 8 Transfer-size=9KB

Shared Bottleneck: RTT 2 18 1 Base RTT=ms 8 7 8 (a) Mean completion time 16 14 12 8 6 4 2 7 8 (b) Retransmissions IETF96 LISA: A Linked Slow-Start Algorithm for 8 / 16

Shared Bottleneck: Bandwidth 5 4 3 2 BW=Mbps Base RTT=ms 7 8 9 (c) Mean completion time 7 8 9 (d) Retransmissions 6 5 4 3 2 BW=45Mbps Base RTT=ms IETF96 LISA: A Linked Slow-Start Algorithm for 9 / 16 18 1 1 1 8

Non-Shared Bottleneck: RTT 1 8 Base RTT=ms BW=5Mbps 8 118 (a) Mean completion time 8 7 8 1 18 (b) Retransmissions BW=5Mbps Base RTT=ms 8 118 8 118 IETF96 LISA: A Linked Slow-Start Algorithm for / 16 2 1

Non-Shared Bottleneck: Bandwidth 5 4 Base RTT=ms 3 2 BW=35Mbps 8 118 (a) Mean completion time 9 8 7 8 1 18 (b) Retransmissions 5 4 3 2 BW=45Mbps Base RTT=ms 35 25 15 5 8 118 8 1 18 IETF96 LISA: A Linked Slow-Start Algorithm for 11 / 16

Shared Bottleneck: Large transfer R ss (in pkts) 18 1 1 1 8 Bandwidth=5Mbps 8 1 1 1 18 Base RTT (in ms) (a) Vary RTT R ss (in pkts) 2 1 Base RTT=ms 5 15 25 Bandwidth (in Mbps) (b) Vary Bandwidth Number of Retransmissions at the end of SS IETF96 LISA: A Linked Slow-Start Algorithm for 12 / 16

Non-Shared Bottleneck: Large transfer 3 3 R ss (in pkts) 2 1 Bandwidth=5Mbps R ss (in pkts) 2 1 Base RTT=ms 8 1 1 1 18 Base RTT (in ms) (a) Vary RTT 5 15 25 Bandwidth (in Mbps) (b) Vary Bandwidth Number of Retransmissions at the end of SS IETF96 LISA: A Linked Slow-Start Algorithm for 13 / 16

Thank you! IETF96 LISA: A Linked Slow-Start Algorithm for 14 / 16

LISA in Shared Bottlenck Server Eth Eth1 Bottleneck Link (a) Shared bottleneck LISA behaves less aggressive than Eth Client Retransmissions sum of cwnd 8 7 Transfer-size=5KB.1.2.3.4.5.6.7.8.9 1 LISA Time (in sec) (b) Total cwnd LISA.1.2.3.4.5.6.7.8.9 1 Time (in sec) Transfer-size=5KB IETF96 LISA: A LinkedTotal Slow-Start retransmissions Algorithm for 15 / 16

LISA in Non-Shared Bottlenck Server Bottleneck Link Bottleneck Link (a) Non-Shared bottleneck cwnd (in packets) 7 Client Retransmissions.1.2.3.4.5.6.7.8.9 1 35 25 15 Time (in sec) main flow subflow 5 LISA main flow LISA subflow.1.2.3.4.5.6.7.8.9 1 main flow subflow LISA main flow LISA subflow Transfer-size=KB Time (in sec) (b) Retransmissions IETF96 LISA: A Linked Slow-Start Algorithm for 16 / 16