I am generally excited about JBoss Web and I hope it gets adopted soon.

Thanks for your feedback,
Ioannis

Nice test, but I would suggest you try that from separate boxes connected with at least 1GBPs network. Like your test are showing the real numbers are pretty random, because both JMeter and JBossWeb are Java applications, so the GC can interfere with the results.

Also, since JBossWeb is semi-event driven it will use less resources the standard Tomcat, so if you use the test client on the same box it will consume more resources.

If interested you can find some benchmark results on:
http://labs.jboss.com/file-access/default/members/jbossweb/freezone/images/benchmarks/10kssl.png
This is 10K binary file served via SSL

http://labs.jboss.com/file-access/default/members/jbossweb/freezone/images/benchmarks/1kssl.png
Same thing for 1K binary file served via SSL

http://labs.jboss.com/file-access/default/members/jbossweb/freezone/images/benchmarks/ka50log.png
This is comparison with Apache Httpd as well. You can see that for large files where OS sendfile is used the performance is 4 times higher the for standalone Tomcat, and equal to Httpd (they both use the same code from APR).

http://labs.jboss.com/file-access/default/members/jbossweb/freezone/images/benchmarks/jio.png
http://labs.jboss.com/file-access/default/members/jbossweb/freezone/images/benchmarks/apr.png
This is CPU and Memory usage for Tomcat and JBossWeb. JBossWeb uses much less resources for static content delivery.

In any case, thanks for keeping an eye on JBossWeb. Currently we are in the process of creating JBossWeb 2.0 based on JBossAS 5 and Tomcat 6.

Regards,
Mladen Turk
JBossWeb Lead
Red Hat, Inc.

thanks for sharing this info and thanks for bringing JBoss Web to our attention.