The C1000K Problem

I've always liked Dan Kegel's C10K page[1], but 10,000 connections is now easily doable on low-end hardware with minimal tuning. As a thought experiment (and maybe a real experiment, depending) I want to try for 1,000,000 connections.

Massive numbers of TCP/IP connections are nice for a number of reasons: Web applications that do a "wait on an XMLHttpRequest" loop need many idle connections, and I'm thinking that any solution to the RSS-polling problem[2] will involve something similiar.

It's not exactly a new idea, so it's going to be as much a web-research thing as a coding thing. Last time[3] I played around with server scalability was loads-o'-fun, I suspect this time will be even more educational. I'll probably end up documenting the results over on Distributopia, it seems more appropriate for longer-form articles.

[1] http://www.kegel.com/c10k.html
[2] Have all the readers connect, then wait. It's push all over again. Wait, it _is_ push all over again. I feel ill.
[3] http://www.distributopia.com/rotd3_archive.html

tags: c10k, dankegel, tcp/ip, scalability, c1000k