More on the microbenchmark

That hasn't changed.

Yesterday I blogged about a small test I made of Orion and Tomcat, and some data that came out of it. The data is interesting, but only to a degree - and I have to admit, I was stupid for putting timing data into the test. The timing data is entirely irrelevant. Nobody really uses HTTP 1.0, and nobody hits an app server like I was hitting it.

The goal was, pure and simple, to reap session identifiers. Past that, the fault list is amazing:

1. I ran the containers with identical settings (-Xmx128m -Xms128m -server -verbosegc, for Orion, Tomcat, Resin, and Jetty - and left WebLogic and WebSphere untested). This means that the servers might not be getting reasonable configurations, which affects speed. The timing results were of secondary interest - and aren't valid!
2. I ran the test for a small sample (100000 entries).
3. Of those 100K entries, the summarization of the session ids was in order BY THREAD, so I don't actually see the variation of session ids as they're generated - only as each individual thread sees them. You could say it breaks down like this:

   Order of Occurrence	Summary Order
   Thread Session 1 AAAAAA 2 AAAAAB 1 AAAAAC 2 AAAAAD	Thread Session 1 AAAAAA 1 AAAAAC 2 AAAAAB 2 AAAAAD

   This is suboptimal, especially when the variance between session ids is larger.

And yes, I'm planning on doing more analysis and exposition.