Go to the first, previous, next, last section, table of contents.

Large Site Load Balancing

Imagine a Muq site based on dozens to hundreds of interchangable server boxes. They might be clustered Beowulf style or WAN-distributed over the Internet, it doesn't matter too much from a design perspective.

At login, the least-loaded machine close to the point of login is selected, the user's dbfiles copied to that machine and mounted (or perhaps accessed via NFS or such) and the login redirected to that machine.

Voila, Muq sites scalable almost indefinitely!

All it really requires is good implementation of dbfile dis/mounting, plus a teensy bit of logic to organize and migrate dbfiles, and of course some simple mechanism for redirecting logins to a selected machine.

(A related idea might be implementing support for distributing datasets over many machines. Suppose the set of Muq packages gets too large to store on a single site, or that one wants to operate on sets or numbers too large to store on one machine. All it should take to make it work are a hashing algorithms to distribute data across the available boxes plus appropriate datastructures and operations to access such distributed data.)

This one will stay on my personal back burner unless/until I run into a personal need for such a facility...

Go to the first, previous, next, last section, table of contents.