Control Program (Overseeing Agents)

Question

Hey guys.
I'm an Artificial Intelligence student at University, working on my final year project. It's actually quite interesting. It's a conceptual indexer for text retrieval. My project involves a lot of 'agents' working together, and I wanted to create a kind of 'master' program to oversee the agents, queue tasks up, make sure that none of agents has crashed or frozen, stuff like that. The part i'm having some trouble with is the crashing and freezing part. I'm not really sure how to even begin going about it. I'm sure it's actually pretty easy, but I was wondering if anyone would be able to make some suggesstions to send me down the right path.
Let me explain a little more about what I would like it to do.
Basically there are five agents working together to create an information retrieval system. One converts HTML and RTF documents to plain text. The next uses WSD techniques to give each word it's correct sense (context). After that the document is conceptualised (given overall and sectional concepts). This information is then indexed. There is also a seperate agent for searching. In order to keep this system efficient and stable, on top of this will be a master agent, queing tasks and making sure that all the other agents are working as they should be. The problem im having is with how to make sure that one of the other agents hasn't crashed or frozen. If it has crashed or frozen, I would like the master agent to shut it down and restart it. My first thought about this was to have some kind of ping system, where the master agent would send a message to the other agents, and if it gets no reply then it restarts them, but there must be a more accurate and efficient way then this surely?
I don't want the actual answer for this, just for someone to show me the right path to follow. It's just that i'm not even sure where to begin with this one.
Anyway, thanks in advance.
Adam.

Answer 1

This is one of those rare situations where the depracted Thread.stop method seems to be unavoidable.
On the face of it crashes are easilly dealt with, but if it's a RuntimeError is there realy much point in restarting the Aglet? A RuntimeError usually implies that the code is faulty.
You'll need to use timeouts to monitor for Aglets that get stuck. For example your Aglet should send some kind of "I'm still here" periodically which resets a timer in a separate thread.
You should study ClassLoaders. It's probably best to load your Aglets from a seperate class directory, jar, or set of same using a URLClassLoader. That enables you to do stuff like "hot" reloading, and start thinking about security policies (especially if your aglets may be writtern by third parties).
I'd recomend the book "Inside the Java2 Virtual Machine".