Server comparison report. The (for pay) report provides a detailed analysis, evaluation and comparison of AppServers from IBM, Fujitsu, SilverStream, BEA, Borland and JBoss. Full results for JBoss are available for free, as well as anonymous cross-server results, and a comparison of ECperf to their own benchmark app.
- Posted by: sharat nellutla
- Posted on: March 04 2002 10:34 EST
read more @ http://www.cmis.csiro.au/adsat/j2eev2.htm.
- CSIRO J2EE Application Servers Report Version 2.0 Available by Guglielmo Lichtner on March 05 2002 09:49 EST
- CSIRO J2EE Application Servers Report Version 2.0 Available by Paul Brebner on March 06 2002 00:33 EST
From the report: "At client loads of around 100 per application server, JBoss provides good stateless session bean performance. Performance however declines as the client loads increase. This means that a JBoss application may not be able to scale to handle very large bursts in traffic."
If anyone knows why this happens I would love to hear it. I think this problem might be alleviated by using tcp/ip multiplexing. Would any jboss insiders care to comment?
We've seen this behaviour with a few of the AppServers in the past. It seems to be caused by resource contention - the higher the concurrency, the slower they go. Most products get around this by allowing the maximum concurrency to be limited - typically by setting max threads in the ORB.