Evaluation of TCP BBR in Comparison with Other Congestion Avoidance - PowerPoint PPT Presentation

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Evaluation of TCP BBR in Comparison with Other Congestion Avoidance Algorithms Benedikt Jaeger September 13, 2017 Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Contents TCP BBR Related Work Parameters & Measurements Next Steps B. Jaeger – TCP BBR 2

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Bottleneck-Bandwidth & RTT Congestion-based Congestion Control 1. estimate bandwidth (max-filter) 2. measure propagation delay (min-filter) 3. compute BDP B. Jaeger – TCP BBR 3

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Bottleneck-Bandwidth & RTT Congestion-based Congestion Control 1. estimate bandwidth (max-filter) 2. measure propagation delay (min-filter) 3. compute BDP Goals: • keep in-flight data = BDP • send with bottleneck-bandwidth B. Jaeger – TCP BBR 3

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Bottleneck-Bandwidth & RTT bottleneck Sending Rate Time B. Jaeger – TCP BBR 4

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Bottleneck-Bandwidth & RTT bottleneck Sending Rate BBR Reno Time B. Jaeger – TCP BBR 5

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Bottleneck-Bandwidth & RTT Wireshark IO Graphs: s1 180000 10 ms Intervals Bandwidth RTT 150000 120000 90000 60000 30000 0 2 4 6 8 10 12 Time (s) Single BBR flow with • 10 Mbit/s bottleneck-bandwidth • 40 ms RTT B. Jaeger – TCP BBR 6

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Related Work Source other mixed different Number CCAs CCAs RTTs of Hosts BBR,2016 x x � � Ma,6.2017 x x � � Hock,8.2017 � ( � ) � � Reproduced Results: BBR,2016: network changes, fairness to itself Ma,6.2017: RTT unfairness Hock,8.2017: wrong operation point (buffers not empty) B. Jaeger – TCP BBR 7

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Mininet Setup Different RTTs Bottleneck Link Sender Receiver B. Jaeger – TCP BBR 8

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Parameters & Measurements Parameters • Congestion-Control algorithm per flow/host • bottleneck-bandwidth & TBF parameters • number of flows & individual RTT Different RTTs Bottleneck Link Sender Receiver B. Jaeger – TCP BBR 9

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Parameters & Measurements Measurements • tcpdump: throughput, RTT, fairness • BBR values: btl_bw, rt_prop, pacing_gain B. Jaeger – TCP BBR 10

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Next Steps Find out how BBR works � Implement Setup in Mininet � Reproduce previous Results in Mininet � Develop tool-chain to interpret Results in progress Find meaningful Test Scenarios in progress Run final Tests Verify Results on Hardware Interpret Results B. Jaeger – TCP BBR 11

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Related Work Sources: N. Cardwell, Y. Cheng, C. S. Gunn, S. H. Yeganeh, and V. Jacob- son. BBR: Congestion-Based Congestion Control. ACM Queue, 14, September-October:20 – 53, 2016. S. Ma, J. Jiang, W. Wang, and B. Li. Towards RTT Fairness of Congestion- Based Congestion Control. M. Hock, R. Bless, and M. Zitterbart. Evaluation of BBR Congestion Control. B. Jaeger – TCP BBR 12

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Intra-Protocol Fairness Wireshark IO Graphs: s1 1.8·10 6 1.5·10 6 1.2·10 6 Bits/100 ms 900000 600000 300000 0 0 15 30 45 60 75 90 Time (s) • 10 Mbit/s bottleneck-bandwidth • 5 BBR flows each with 40ms RTT starting at intervals of 2 s B. Jaeger – TCP BBR 13

Chair of Network Architectures and Services Department of Informatics Technical University of Munich BBR vs CUBIC Wireshark IO Graphs: s1 1.5·10 6 1.25·10 6 1·10 6 Bits/100 ms 750000 500000 250000 0 0 20 40 60 80 100 120 Time (s) • CUBIC, BBR, 30ms RTT, 10Mbit/s • TBF latency 300ms (large buffer) B. Jaeger – TCP BBR 14

Chair of Network Architectures and Services Department of Informatics Technical University of Munich BBR vs CUBIC Wireshark IO Graphs: s1 1.05·10 6 900000 750000 Bits/100 ms 600000 450000 300000 150000 0 0 20 40 60 80 100 120 Time (s) • CUBIC, BBR, 30ms RTT, 10Mbit/s • TBF latency 50ms (small buffer) B. Jaeger – TCP BBR 15

Chair of Network Architectures and Services Department of Informatics Technical University of Munich BBR vs CUBIC Wireshark IO Graphs: s1 1.2·10 6 1·10 6 Bits/100 ms 800000 600000 400000 200000 0 0 20 40 60 80 100 120 Time (s) • CUBIC 30ms RTT, BBR 80ms RTT, 10Mbit/s • TBF latency 50ms (small buffer) B. Jaeger – TCP BBR 16