slide1
Download
Skip this Video
Download Presentation
Cheng-Hsin Hsu, Nitin Chiluka, and Mohamed Hefeeda School of Computing Science, Simon Fraser University, Canada

Loading in 2 Seconds...

play fullscreen
1 / 1

Cheng-Hsin Hsu, Nitin Chiluka, and Mohamed Hefeeda School of Computing Science, Simon Fraser University, Canada - PowerPoint PPT Presentation


  • 103 Views
  • Uploaded on

ISP-Friendly Peer Matching. Cheng-Hsin Hsu, Nitin Chiluka, and Mohamed Hefeeda School of Computing Science, Simon Fraser University, Canada. 1. Motivation. 2. Big Picture. P2P costs ISPs more money Challenge: find senders to reduce loads on inter-ISP links improve application performance

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Cheng-Hsin Hsu, Nitin Chiluka, and Mohamed Hefeeda School of Computing Science, Simon Fraser University, Canada' - howe


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
slide1

ISP-Friendly Peer Matching

Cheng-Hsin Hsu, Nitin Chiluka, and Mohamed Hefeeda

School of Computing Science, Simon Fraser University, Canada

1. Motivation

2. Big Picture

  • P2P costs ISPs more money
  • Challenge: find senders to
    • reduce loads on inter-ISP links
    • improve application performance
  • Solution: ISP-friendly matching
    • find senders to minimize AS distance
    • within AS, get closer senders by IP prefix
  • Match senders online using oracle
    • faster distance lookup
    • smaller data structure in memory
    • exact/approximate distance
  • Infer AS distance offline distance oracle
    • leverage public info
    • efficient inference algorithm

3. Our Approach

  • V : all ASes; L : stub ASes  Most ASes are in L (87%)
  • Construct_Distance_Oracle
  • For , compute shortest up-hill distance
  • For , compute valley-free distance
  • For and , compute valley-free distance
  • For , compute valley-free distance
  • Compute shortest valley-free AS paths
    • valley-free: customer AS does not transit data for its providers
  • Current algorithms [e.g., Mao 05]

 time

    • runs in ~ 2 days (25,000+ ASes)
    • needs ~ 625 MB memory
  • Our proposed algorithm
    • preprocess AS graph  concise data structure: Core Matrix
    • exclude stub ASes; they don’t transit traffic for any other ASes
    • runs in ~ 3 hours
    • needs ~ 10 MB memory (1.7 %)

Step 1

Step 2

Step 3

Determine distance for stub t is

a couple of comparisons

Only Core matrix is stored

 Small memory footprint

5. Work-in-Progress

4. Sample Results

  • Improve accuracy of AS distance inference (~70 %)
  • Even more compact data structure
    • recursively remove stub ASes
  • Real implementations and measurements
    • BitTorrent, peer-assisted CDN
  • http://nsl.cs.sfu.ca/wiki
  • For AS relationship

info from CAIDA on

2008/06/23

  • Client IPs:
    • 147, 822 from BitTorrent trackers
    • 160, 543 from CBC/Radio-Canada servers

CBC, 2000 Potential Senders

BitTorrent, 10 Senders

Up to 15times reduction on Inter-ISP traffic

ad