broadband(internetperformance:((...

Broadband Internet Performance: A View from the Gateway

Srikanth Sundaresan, Walter de Donato, Nick Feamster, Renata Teixeira, Sam Crawford, Antonio Pescapè

Georgia Tech University of Napoli

CNRS/UPMC Sorbonne Universités SamKnows

What Affects Broadband Performance?

•  Important for regulators, consumers, ISPs •  NoQon of performance is fuzzy – What metrics should we measure?

– How to measure them? Broadband Internet Performance: A View from the Gateway ACM SIGCOMM, Toronto., August 2011 2

Accurate Measurements are Difficult

End host measurements are not con/nuous, and affected by confounding factors

Home Network: AT&T DSL 6 Mbps Down, 512 Kbps Up

Last Mile ISP Network

speedtest.net: 4.4 Mbps, 140 Kbps

Netalyzr: 4.8 Mbps, 430 Kbps

Broadband Internet Performance: A View from the Gateway ACM SIGCOMM, Toronto., August 2011 3

The Case For the Gateway

Gateway enables periodic measurements, and can account for confounding factors

Home Network: AT&T DSL 6 Mbps Down, 512 Kbps Up

Last Mile ISP Network

speedtest.net: 4.4 Mbps, 140 Kbps

Netalyzr: 4.8 Mbps, 430 Kbps

Gateway: 5.6 Mbps, 460 Kbps


ContribuQons

•  Study of access networks from the gateway •  Insight into how to benchmark performance – Effect of measurement technique, traffic shaping

– Effect of access link properQes, home equipment

•  Insight into broadband performance – Consistency of performance

– Loss, ji`er, Qme-‐of-‐day effects


The Deployments

•  Breadth: The FCC/SamKnows study –  4,000 gateways, 16 ISPs, mulQple service plans

•  Depth: The BISmark study –  16 gateways in Atlanta, on-‐demand measurements

•  DuraQon: Dec 2010 – Jan 2011

SamKnows/BISmark


Last Mile Internet

Nearby Server

Results: Overview

•  Throughput measurement technique depends on usage scenario

•  Traffic shaping is highly variable across users

•  Access link characterisQcs affect performance

•  Modem buffers induce high latency


Results: Overview






InterpreQng Throughput Results

Different techniques measure different aspects of throughput


Results: Overview






Traffic Shaping: PowerBoost

•  Cable companies adverQse “PowerBoost” – Short bursts of high bandwidth

•  Non-‐existent in DSL


Traffic Shaping Varies Across Users

Short-‐term throughput significantly different from sustainable throughput


Results: Overview






Latency Measurements


Parameter What it captures

End-‐to-‐end Latency to nearby server

Last-‐mile Latency to edge of ISP network

Under Load Buffer delays due to cross traffic

DSL last-‐mile latencies can be very high

Impact of Last-‐mile on Latency


Cable ISPs DSL ISPs

DSL Interleaving Affects Latency

9 8 7 6 5 4 3 2 1

Fastpath is suscepQble to bursty loss

9 6 3 8 5 2 7 4 1

Fastpath sends data in order, can recover from single losses

Interleaving sends data out-‐of-‐order, can recover from bursty loss

Fastpath vs. Interleaved last-‐mile data path


Extra Delay

Example: Latency-‐Throughput Tradeoff


User 2: high latency, low loss

User 1: low latency, high loss

Throughp

ut (M

bits/s)

Latency (ms)

10

2.5

2.2

30

Both users have same service plan

Interleaving decreases loss, increases latency, improves throughput

Interleaving creates a trade-‐off between latency and throughput

Results: Overview






Modem Buffers are Too Large

Service plans can interact badly with modem buffers


10 seconds!

Conclusion

•  The gateway provides unique insight into home network

•  Throughput measurements are affected by measurement technique, shaping

•  Latency is affected by last-‐mile, buffering

•  ISPs generally deliver consistent throughput, with some Qme-‐of-‐day effects

•  How to miQgate buffering effects


Future Work

•  BISmark node deployment across the world – Currently nodes in US, South Africa

•  Performance monitoring from the edge – Effect of peering on performance –  IPv6 performance – Effect of CDN locaQon, traffic engineering on applicaQon performance

•  Measure the home network


h^p://projectbismark.net


Previous Studies

•  Measurement from outside – Dischinger et al [1] iniQated measurements from wide area server

–  Indirect measurements, not conQnuous

•  Measurements from inside –  Endhost based measurements

–  Grenouille [2], Netalyzr [3] –  Can’t account for confounding factors

[1] Characterizing ResidenQal Broadband Networks IMC 2007 [2] h`p://www.grenouille.com [3] h`p://netalyzr.icsi.berkeley.edu Netalyzr: IlluminaQng the Network Edge IMC 2010


broadband(internetperformance:((...

Documents