Devloping a chat application in python

Question

I want to develop a chat application similar to omegle.com.

FUNCTIONALITY This website picks up two random people who are logged into the website and then pairs them for a chat session. Now if one of the party A disconnects then the other party B is informed and then the party B is connected to a different party. I want my application to handle heavy load and it should not crash.

TECHNICAL: Now I had a few doubts about the architecture of the application:

1. Should every message sent by A or B go through the server? Will it cause loading of the server? OR should the server be used only for establishing the connection between the two parties?

2. I have heard that the omegle.com is developed using python using the Twisted library. I also heard that supports easy scaling and can handle huge loads. If any one has experience with the Twisted framework I would like to hear their comments.

Answer 1

You can set up some form of P2P connections between your users or you can do it through Servers. It all depends on what your interface is going to be. If its going to be in a browser then I would recommend routing it through a server and just having the some AJAX app that poles for changes.

if you worried about scalability why not write your app to work on Google App Engine and that will take care of a lot of your scalability issues and you just write for the platform.

Answer 2

1. Well, you can't send the message between the two parties specifically, unless you use an appropriately-authorised clientside system (Signed Java Applet, etc), so the answer here is obvious.

-- Edit:

By the way, a trick, for perhaps someone 'new' to programming; if you want to develop an application that "scales" I suggest to you that you not only research what, exactly, you mean by "scaling", but also just implement something, and then see where it goes wrong. It can be fun (arguably) to implement it yourself and find out the issues that other people talk about, with direct experience. You can set up systems to determine the 'load' (in whatever context you determine) it can handle, and then you can determine how to do it better. It is arguably wasted time, if you need to re-write, but perhaps it's valuable. Personally, (having done that already, to a degree) I'd go the research and planning route :)

Answer 3

1. Every request to server cause another load. The best is to pair clients without server. Alternative is to use dedicated light server for handling asynchronous request to avoid overload main website.

2. About Twisted, everything depends on... For example - my company heavily used twisted for multimedia (streaming and desktop applications). From second side - you can try tornado http://github.com/facebook/tornado or just erlang language http://erlang.org.