The Performance Benefits of Multihoming - PowerPoint PPT Presentation

About This Presentation
Title:

The Performance Benefits of Multihoming

Description:

Avg. Performance of random set of k ISPs compared best set of k. 1. Random choice ... active clients: download 50K objects every 6 min ... – PowerPoint PPT presentation

Number of Views:22
Avg rating:3.0/5.0
Slides: 2
Provided by: jon793
Category:

less

Transcript and Presenter's Notes

Title: The Performance Benefits of Multihoming


1
The Performance Benefits of Multihoming A.
Akella, B. Maggs, S. Seshan (CMU), A. Shaikh (IBM
Research), R. Sitaraman (Akamai)
1. Goals
2. Approach
Popular content providers
Traditional Multihoming
Multihoming today
Web server
Either path can be used
Destination
Destination
Primary
Back-up
Internet
Internet
Enterprise
Active clients
Goal Optimize download performance
Goal Optimize client-perceived download
performance
AS 300
AS 200
AS 300
AS 200
4.0.0.0/19 AS-path 101
4.0.0.0/19 AS-path 101 101 101
Web server
Stand-ins for multihomed webserver
metro areas
Route-control
AS 1014.0.0.0/19
Use AS200 or 300?
AS 101
  • Primary motivation Reliability
  • Use in the primary- backup mode
  • Multihoming for Performance
  • Performance varies with time, ISPs
    choose routes cleverly

Akamaiservers
Stand-ins foractive clients download 50K
objects every 6 min
For an enterprise or a content provider in
a metro area
Recruit Akamai servers represent multhihomed
webserver and its clients 1. Wide-spread
geographic diversity 2. In each city, servers
connected to variety of ISPs ? Helps do a
thorough study
  • The maximum performance benefits
  • Realizing these benefits in practice

3. Performance Benefits
4. Practical Multihoming
k-multihoming Performance
Assumed perfect information and instantaneous
route changes so far ? impractical1. Use history
to track performance EWMA(a)2. Limit route
changes update every T mins
Performance of best set of k ISPs compared with
all ISPs 1. Lots of benefit from using
multiple ISPs 2. Benefits beyond 4 providers
marginal Avg. Performance of random set of k
ISPs compared best set of k 1. Random choice
could perform 50 or worse 2. Important to
carefully choose ISPs
a1, T30 minutes
a10, T30 minutes
Performance of a Random Choice
1. a1 is clearly better2. Performance could
get worse with more ISPs!!? should give
recent samples more weight? should collect
timely and accurate samples
Write a Comment
User Comments (0)
About PowerShow.com