Arranging data

Swarm must find a way to adaptively arrange data on servers to minimize the number of times a thread must jump from one computer to the other.

Ian has posted a proposal for this to our mailing list, and more recently done some initial work on simulation. View the proposal for an algorithm here, and if interested - please join the conversation!

Concurrency

Ian proposed examining "software transactional memory" as a way to prevent threads from interfering with each-other in bad ways. This spawned a discussion on the pros and cons of various approaches. Read the thread here.

Universal code access

Swarm requires that every Swarm node be running the same codebase.

I think this can be achieved by hooking into the JVM's classloader and making it retrieve packages on-demand as they are needed. This concept is inspired by the Zero Install package management system.

This would probably involve having some kind of public crypto key in the package name as a unique identifier - eg:

package swarm.bR781Z1h7.wombat.3.0