Remote Locking

Question

I am designing a remote threading primitive protocol. Currently we only needs mutexes (i.e. Monitors) and semaphores. The main idea is that there doesn't need to be a central authority - the primitives should be orchestrated amongst the peers that are interested in them.

I have bashed a few ideas around on paper and in my head for a few weeks; but I think I should really have a look at prior literature. Is there any?

It will run over XMPP - but that is an implementation detail. I am just looking for a specification or such on the actual protocol flow - so it doesn't really matter what protocol the literature originates from.

Thanks a million.

Answer 1

Distributed mutexes are tricky structures. You need to handle all sort of weird conditions that do not exist with a single machine based implementation. In particular you need to handle situations where agents lose comms with the group and still hold a lock on a shared resource. Additional to that scenario there are complex scenarios where your group is fractured and you grab a lock on a resource. When the fractured groups join into a big group you need some way to reconcile the locks which is far from trivial.

I would strongly recommend looking into some messaging based middleware such as Erlang and JBoss

I would also recommend posting separate questions on the particular distributed algorithms / data structures you need implemented. It's possible that you could get away with an out-of-the-box implementation in a middleware library that could be adjusted to meet your needs.