bismark caida 022011

Upload: prasanta-sanyal

Post on 13-Apr-2018

232 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/26/2019 Bismark Caida 022011

    1/24

    Benchmarking Broadband Internet

    Performance

    Srikanth Sundaresan, Walter de Donato,

    Nick Feamster, Renata Teixeira, AntonioPescape

  • 7/26/2019 Bismark Caida 022011

    2/24

    What is the Performance of NetworkAccess Links?

  • 7/26/2019 Bismark Caida 022011

    3/24

    Previous Performance Studies

    Study from outsideDischinger et al. (IMC 2008), Netalyzr (IMC 2010)Not continuous, not many per user, no view into home

    Study from insideGrenouille projectHard to account for device diversityHard to account for home network

  • 7/26/2019 Bismark Caida 022011

    4/24

    The BISMark Project

    Periodic measurements to last mile and end-to-endMeasure directly at the gateway deviceAdjust for confounding factors

  • 7/26/2019 Bismark Caida 022011

    5/24

    BISMark

    Deploy programmable gateways in homesNoxBox deployment: about 35 around AtlantaSamKnows deployment: about 10000 around the

    US

    NoxBox Netgear

  • 7/26/2019 Bismark Caida 022011

    6/24

    Gateway Vantage Point: Advantages

    Observes all traffic passing through network

    Isolate individual factors affecting networkperformanceWirelessCross trafficLoad on measurement host

    End-to-end pathConfiguration

  • 7/26/2019 Bismark Caida 022011

    7/24

    Current Deployment

    16 boxes deployed

    10 in ATT, 4 in Comcast, 2 ClearWire

    Most of the deployments within Atlanta

    All measurements done to server at Georgia

    Tech

  • 7/26/2019 Bismark Caida 022011

    8/24

    Active Measurements

  • 7/26/2019 Bismark Caida 022011

    9/24

    Results

    ThroughputDifferent throughput techniques capture differentaspects of throughputThere is high variation across users with same

    technique

    LatencyLatencies vary within the same ISP

    Last-mile latencies are significantModem buffers are too largeModifying data transfer using using traffic shapingmight mitigate the problem in the short term

  • 7/26/2019 Bismark Caida 022011

    10/24

    Different Techniques, DifferentAspects of Throughput

    Single threaded is what users see on a singledownloadWeb browsing is mostly multi-threaded

  • 7/26/2019 Bismark Caida 022011

    11/24

    Different Users, Different Performance

    Same service plan & ISP, different loss profileUser 1 sees much more loss, but also muchlower latency

    User 2 has interleaving turned on

  • 7/26/2019 Bismark Caida 022011

    12/24

    Traffic Shaping DiffersAcross Users

    Different burst magnitudesDifferent lengths of time

    Download shaping

  • 7/26/2019 Bismark Caida 022011

    13/24

    Traffic Shaping under Upload

    How do we account for such variance?Implications for speed test results?

  • 7/26/2019 Bismark Caida 022011

    14/24

    Results

    ThroughputDifferent throughput techniques capture differentaspects of throughputDepending on how throughput measurements are

    conducted, they may vary considerably across users

    LatencyLatencies vary within the same ISP

    Last-mile latencies are significantModem buffers are too largeModifying data transfer using using traffic shapingmight mitigate the problem in the short term

  • 7/26/2019 Bismark Caida 022011

    15/24

    Last mile latency variesacross users

    RTT(ms)

    RTT(ms)

    Baselines Different for 2 ATT customers.Same service plan, within a few blocks of each other.

    Interleaving modes are different.

  • 7/26/2019 Bismark Caida 022011

    16/24

    Last-Mile Latencies are Significant

    All but 2ms comes from last mile

    High correlation (0.95) with end-to-end latency

    End-to-endlatency

    Last milelatency

  • 7/26/2019 Bismark Caida 022011

    17/24

    Effect of Access Link Technology

    Baseline latency dependent on accesstechnologyADSL last mile 8 to 25ms, Comcast ~10msWiMAX ~ 75ms!

    RTT(ms)

    RTT(ms)

    Comcast Clear

  • 7/26/2019 Bismark Caida 022011

    18/24

    Buffers are Too Large

    Buffering in modems can be as high as ten seconds!Can be empirically modeled with token-bucket filter

    Also exist elsewhere in the stack

    Latency profile while saturating upstream link

  • 7/26/2019 Bismark Caida 022011

    19/24

    Traffic Shaping Affects Latency, Too

    After different periods of time, latency and lossprofiles change dramatically

  • 7/26/2019 Bismark Caida 022011

    20/24

    and in different ways

    Possible cause: dynamic buffer sizing

  • 7/26/2019 Bismark Caida 022011

    21/24

    Keeping Latency Under Control

    Intermittent or shaped traffic can achieve samelevels of throughput, without incurring high latency

  • 7/26/2019 Bismark Caida 022011

    22/24

    Other fixes for Latency Under load

    Shaping traffic comes at the cost of sacrificingthroughput

    Is it possible to fix latency without affecting

    throughput?Smaller buffers might affect long flowsSome sort of Active Queue Management?

    RED, Fair queueing

  • 7/26/2019 Bismark Caida 022011

    23/24

    Takeaway Lessons

    One measurement does not fit allDifferent measurements yield different resultsDifferent ISPs have different shaping behaviors

    One ISP does not fit allThere is no best ISP for all usersDifferent users may prefer different ISPsThere is a need for a nutrition label

    Home network equipment can significantly affect

    performance

  • 7/26/2019 Bismark Caida 022011

    24/24

    Thanks!

    Comments?

    [email protected]