Proper way to implement a Direct Connect client in Twisted?

Question

I'm working on writing a Python client for Direct Connect P2P networks. Essentially, it works by connecting to a central server, and responding to other users who are searching for files.

Occasionally, another client will ask us to connect to them, and they might begin downloading a file from us. This is a direct connection to the other client, and doesn't go through the central server.

What is the best way to handle these connections to other clients? I'm currently using one Twisted reactor to connect to the server, but is it better have multiple reactors, one per client, with each one running in a different thread? Or would it be better to have a completely separate Python script that performs the connection to the client?

If there's some other solution that I don't know about, I'd love to hear it. I'm new to programming with Twisted, so I'm open to suggestions and other resources.

Thanks!

Answer 1

Without knowing all the details of the protocol, I would still recommend using a single reactor -- a reactor scales quite well (especially advanced ones such as PollReactor) and this way you will avoid the overhead connected with threads (that's how Twisted and other async systems get their fundamental performance boost, after all -- by avoiding such overhead). In practice, threads in Twisted are useful mainly when you need to interface to a library whose functions could block on you.