I've been looking at metrics - currently the model is multiple machines, connecting to a central metrics server that stores metrics in postgres. I'm going to need something like that, so that's good, but it should also scale downwards. Oh, and Sean, I couldn't actually get values added by test-client.py to show up in the web interface - maybe try recreating your test environment from scratch and see what step are missing from Twisted? I keep getting these errors: 26/12/2001 16:16 [Broker,0,127.0.0.1] ERROR: unknown variable counter3 So, what metrics needs: 1. Ability to store metrics in memory or in a round-robin database, and on the local server too (for smaller setups and for easier testing.) See the Round Robin Archives section of http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/manual/index.html for details on what it is - it's a more reasonable way of storing data than a RDBMS for this sort of thing in a lot of cases. 2. OS statistics for Unix - does resource.rusage() give the necessary info? 3. Docs, explanations, examples of protocol specific statistics (e.g. Web). 4. An additional way of keeping track of open connections and their metrics, which should combine with the current metrics system. 5. Exporting stats to Windows NT Performance Monitor so it can do pretty graphs and stuff.