SPAND: Shared Passive Network Performance Discovery

Project overview | People | Publications | Related work | Software | Demos! | Email Lists

Project Overview

In the Internet today, it is difficult to determine in advance what the network performance between a pair of internet hosts will be. There are several examples of applications that could use this information if it were made available to them:

1. Clients that must choose between servers that replicate the same content.

2. Servers that present clients with differences in content fidelity of web objects.

3. Applications that present feedback to the user about the quality of connectivity.


SPAND is a software toolkit that makes it easy for networked applications to report the performance they percieve as they communicate with distant internet hosts and remember this information for later use. As clients communicate with other hosts, they generate Performance Reports that summarize how their connection progressed. For example, clients may report the average throughput of a TCP connection or the download time for a particular web page. These Performance Reports are sent to a per-domain Performance Server who acts as a repository of all the Performance Reports for clients in the domain. Later, other clients contact the Performance Server with a Performance Query , asking about the performance to a distant network site. The Performance Server replies with a Performance Response which gives the average performance (and variance of performance) seen by clients in this domain.

The key advantages of this approach are: Take a look at the publications section for more detailed information about SPAND.

People

Publications

Related Work

The IDMaps project at the University of Michigan has a good collection of pointers to other work.

Software

A source distribution of the SPAND toolkit is available. Here is the latest source distribution as a compressed tar file. You may also grab: You can browse the documentation for the latest revision of SPAND here .

Demonstration

Go here to browse the contents of the Performance Server's database. You must have a Java-enabled browser for this page to work.

Email Lists


Maintained by Mark Stemm stemm@cs.berkeley.edu