In another HP sponsored ECperf posting, BEA Weblogic Server 7.0 beta results have been posted that beat both IBM's results and BEA's last published results on the HP LT6000r U3 system. The reason for the performance and price/performance boost: Using the Weblogic JDriver instead of Oracle's JDBC driver.

Check out https://ecperf.theserverside.com.

Okay - this is what vendors should be doing - show us how it performs with different tweeks. Not to be cynical, but I wonder if these results would have been published had the results been slightly less than the first.

Anyway - I hope BEA and other vendors continue to present results based on tweeks and hardware scenarios.

Cheers
Ray

Agreed, Ray. In fact, I'd love to see postings of performance on much smaller systems that couldbe viable for projects with smaller budgets. It's certainly valuable to know where the top end scalability can take you, but I'd be curious to know how given application servers perform on, say, single and dual processor intel machines, as well. The more postings per configuration, the better.

Geoffrey your point highlights the risk that on less impressive hardware these applicationn servers may an unaccepable overhead for their own internals, which could be quit a pickle for medium sized projects...

is there a legal way to use JDriver outside/without Weblogic ?

Oh, okay, is this Performance benchmarking or "benchmarketing"? (;-)
Now, all this fuss about different realizations of J2EE being different in performance characteristics exhibited because of "tweeks" indigenous to a particular vendor's implementation of J2EE and or product suite component, suggests to me, sitting in my garage, what gives then about portability?

They are tweaks to the vendor setup not to the actual codebase so no portability is lost. All of the tests are using the same code, that is the entire point of ECPerf.

Yes, you can buy it separately from BEA ;-)

"I'd love to see postings of performance on much smaller systems that couldbe viable for projects with smaller budgets"

I agree with this too. But the problem is that the ecperf model application is pretty big (an entire supply chain.) So maybe for smaller budgets they should consider making a new application model: a simple application as one typically runs on smaller hardware.

One application I would love to see benchmarked is the "message board" or "discussion thread" (like the one you are reading now). It's a good benchmark app because it has no session data, all the interesting data (the message s) is seen by all users, and therefore the database becomes the bottleneck. This would force vendors to use a large cluster with a distributed cache.

Incidentally, I think that BEA used a budget of about $300K because they wanted to prove that they are more cost-effective than IBM, so they used the same budget as IBM's benchmark.

Guglielmo

The disclosure reports and archive are excellent sources for weblogic tuning information. I can't imagine a more well tuned app than one that the vendors would have to use to compete with each other.

CSIRO (Australian government research organisation) has been conducting extensive testing of middleware technologies for over 2 years. We use a realistic application called "Stock-0nLine" which has been implemented in many different technologies, including J2EE - it's lightweight compared to ECPerf (about 1/3 of the load on the database). We use a level playing field for testing (same h/w, OS, database, etc), and version 2.0 of our J2EE AppServer report is available now.

Check out

http://www.cmis.csiro.au/adsat/j2eev2.htm

The free summary of version 1.0 is also still available at:

http://www.cmis.csiro.au/adsat/mte_reports.htm

Regards,

Paul.

That's nice. I remember years ago when people would buy WebLogic's JDBC drivers because Oracle's were so buggy. It's nice to see they haven't lost the touch.