Distributed Clustering

The Distributed Clustering area aims to solve several problems that currently limit the scalability of web sites as well as cause inefficient use of network resources. The main problems are:

How do you find which web server is "closest" in the network to the client?
How should the "best" web server to serve the client be defined?
What is the best way to load-balance a local cluster of web servers?

A system called WALRUS solves several of these problems and supports BSDI unix.
Another system, Backhand, developed here at CNDS specifically tackles load-balancing local clusters and is available at www.backhand.org. It is a loadable module for The Apache Web Server that allows dynamic request redirection within a cluster of Apache web servers. Redirection is based off various bits of information including, but not limited to, request type and current available resources within the cluster.

Seamlessly Selecting the Best Copy from Internet-Wide Replicated Web Servers
- A research paper on techniques to select which web server a client should automatically be refered to in order to receive the best service on their request. The techniques take into account such factors as network distance, server availability and server load.
- A powerpoint presentation of this paper at the DISC98 conference.
Walrus: Wide Area Load Rebalancing User-transparent System
A high availability load balancing system for the World Wide Web.
Presentation on the Shared-IP Technique for discovery of the closest server
A presentation describing one technique for finding which web server is closest to the client according to network topology.

E-mail webrep@cnds.jhu.edu for more information.

Questions or comments to:

webmaster@cnds.jhu.edu

TEL: (410) 516-5562
FAX: (410) 516-6134

Center for Networking and Distributed Systems
Computer Science Department
Johns Hopkins University
3400 N. Charles Street Baltimore, MD 21218-2686