On the Fairness of Transport Protocols in a Multi-Path Environment

Transcription

1 University of Duisburg-Essen On the Fairness of Transport Protocols in a Multi-Path Environment Hakim Adhari, Martin Becke, Thomas Dreibholz Networking Technology Group Institute for Experimental Mathematics University of Duisburg-Essen

2 Introduction and motivation In the past Single-path dominated network Fairness restrictions TCP-friendly Actually Multi-path transfer is becoming more and more important Congestion Control (CC) assuring Effective use of the resources Fairness But what is fair here? Hakim Adhari, Page 2

3 Multi-Path Congestion Control Multi-Path CC Three goals are targeted [WH9] () Improve throughput (2) Do no harm (3) Balance congestion Considered mechanisms Multi-Path eno (eno-mp) CMT-esource Pooling Version 2 (P-MP-v2) [DBA] Multi-Path TCP (MPTCP) Hakim Adhari, Page 3

4 What I want to talk about today Problem statement Are these mechanisms TCP-friendly/fair? Is fair == TCP-friendly? Approach Analysis of the behavior of the CC Mechanisms Bottleneck topology Disjoint paths topology Hakim Adhari, Page 4

5 Bottleneck Topology Test Scenario F S F F ρ(α) D F S F D Hakim Adhari, Page 5

6 Bottleneck Topology Link-Centric Flow Fairness Expected behavior Goal 2 (Do no harm) [WH9] Link-Centric Flow Fairness ratio of flows to sub-flows : :x For n different flows sharing a link l ρ(l) n for each of the n flows F S F F ρ(α) D B = B + B = ρ(α) 2 F S F D B = B = ρ(α) 2 Hakim Adhari, Page 6

7 Bottleneck Topology Simulation esults ENO-MP 2/3 vs /3 P-MP-v2 and MPTCP /2 vs /2 The mechanisms behave as expected Hakim Adhari, Page 7

8 Disjoint Paths Topology Test Scenario F ρ(α) 2 F S F 3 ρ(β) 4 D F S F D ρ(α) = 2 Mbit/s Hakim Adhari, Page 8

9 Bottleneck Topology Network-Centric Flow Fairness Expected behavior Link Centric Flow Fairness Is this fair? F ρ(α) 2 Link-Centric Flow Fairness? ratio of flows to sub-flows : :x For n different flows sharing a link l ρ(l) n for each of the n flows F S F 3 ρ(β) 4 D B = ρ(α) B = ρ(β) 2 F S F D B = ρ(β) 2 Hakim Adhari, Page 9

10 Bottleneck Topology Network-Centric Flow Fairness Expected behavior Goal 3 (Balance congestion)[hw9] Network-Centric Flow Fairness The whole network is considered F ρ(α) 2 F S F 3 ρ(β) 4 D B = ρ(α) B = max {, ρ β B 2} F S F D B =ρ(β) - B Hakim Adhari, Page

11 Disjoint Paths Topology Simulation esults P-MP-v2 and MPTCP are approximating the Network- Centric Flow Fairness First converging to the fairness line For higher dissimilarity: the single-path flow acquires more bandwidth than the multi-path flow ρ(α) = 2 Mbit/s Hakim Adhari, Page

12 Disjoint Paths Topology Simulation esults Th MP_Flow < Th SP_Flow ρ(α) = 2 Mbit/s Hakim Adhari, Page 2

13 Is this fair what we actually call fair? Example A consumer S (paying for two high-speed Internet access lines) A consumer S (paying for one high-speed Internet access line) Is it really fair if S gets more resources than S? Is it really fair if S gets as much resources as S? - Consumer view - Provider view F S F F 3 ρ(α) ρ(β) 2 4 D F S F F ρ(α) D F S F D F S F D Hakim Adhari, Page 3

14 Is this fair what we actually call fair? Link-Centric Sub-Flow Fairness based on the number of the sub-flows on a link l ρ(l) m for each of the m sub-flows F S F F ρ(α) D B = B + B = 2*ρ(α) 3 F S F D B = B = ρ(α) 3 Hakim Adhari, Page 4

15 Summing up Views on fair esource Allocation We define three different views [BDA2] Link-Centric Sub-Flow Fairness based on the number of the sub-flows on a link l ρ(l) m for each of the m sub-flows Link-Centric Flow Fairness ratio of flows to sub-flows : :x For n different flows sharing a link l ρ(l) n for each of the n flows Network-Centric Flow Fairness the whole network is considered Hakim Adhari, Page 5

16 Disjoint Paths Scenario esults Until now: Dominating single-path transfer Multipath CCs Fair coexistence of multi- and single-path flows Extension of the single-path CC No relationship between the brought-in resources and the final allocation Disadvantage for Multi-Path flows More resources for single-path than multi-path flows With multi-path transfer Are the three goals defined in [WH9] enough? Is an extension / adaptation / correction of the goals needed? Hakim Adhari, Page 6

17 How Should the esources be allocated? Link Network B α F a 2 S F b B β D F c 3 4 S2 F d D2 B γ F e 5 6 S3 D4 Hakim Adhari, Page 7

18 Network-Centric / Link-Centric Sub-Flow Which factors are important to reach the Network-Centric Flow Fairness? Bandwidth? Loss rate? MTU? Delay?.. is it possible to reach the Network-Centric Flow Fairness? What is about the Link-Centric Sub-Flow Fairness? Is the bottleneck problem so important? Hakim Adhari, Page 8

19 Any Questions? Our Team Hakim Adhari Martin Becke Thomas Dreibholz Multi-path transfer project page Hakim Adhari, Page 9

20 eferences [BDA2] Martin Becke, Thomas Dreibholz, Hakim Adhari, and Erwin Paul athgeb. On the Fairness of Transport Protocols in a Multi-Path Environment. In Proceedings of the IEEE International Conference on Communications (ICC), Ottawa/Canada, June [WH9] Costin aiciu, Mark Handley, and Damon Wischik. Practical Congestion Control for Multipath Transport Protocols. Technical report, University College London, London/United Kingdom, [DBA] Thomas Dreibholz, Martin Becke, Hakim Adhari, and Erwin Paul athgeb. On the Impact of Congestion Control for Concurrent Multipath Transfer on the Transport Layer. In Proceedings of the th IEEE International Conference on Telecommunications (ConTEL). [DAB2] Thomas Dreibholz, Hakim Adhari, Martin Becke, and Erwin Paul athgeb. Simulation and Experimental Evaluation of Multipath Congestion Control Strategies. In Proceedings of the 2nd International Workshop on Protocols and Applications with Multi-Homing Support (PAMS), Fukuoka/Japan, March [ADB+] Hakim Adhari, Thomas Dreibholz, Martin Becke, Erwin Paul athgeb, and Michael Tüxen. Evaluation of Concurrent Multipath Transfer over Dissimilar Paths. In Proceedings of the st International Workshop on Protocols and Applications with Multi-Homing Support (PAMS), Singapore, March 2. [DB] Thomas Dreibholz, Martin Becke, Erwin Paul athgeb, and Michael Tüxen. On the Use of Concurrent Multipath Transfer over Asymmetric Paths. In Proceedings of the IEEE Global Communications Conference (GLOBECOM), Miami, Florida/U.S.A., December 2. [Dre2c] Thomas Dreibholz. Evaluation and Optimisation of Multi-Path Transport using the Stream Control Transmission Protocol. Habilitation treatise, University of Duisburg-Essen, Faculty of Economics, Institute for Computer Science and Business Information Systems, March Hakim Adhari, Page 2