Emil Briggs (me@nowherespam.com) on 5/23/06 wrote:
---------------------------
>The Woodcrest and Dempsey numbers used the BEA Jrockit 5.0 >while the Opteron number
>you quoted of 51,106 used the Sun JRE. That makes a direct >comparison meaningless.
>The Opteron number of 62,074 listed on spec.org used the BEA >Jrockit though so a
>direct comparison makes more sense in that case. Doing so >changes the conclusions
>considerably since a 2.4Ghz Opteron system handily beats a >3Ghz Woodcrest.

So, here's how I came to that conclusion. Use the configurable search form in JBB2005.

Search for all systems with:

1. 4 cores
2. A single JVM instance
3. AMD processor

The key point is #2. You cannot compare between systems with multiple JVMs and a single JVM...the multiple JVM always scales much much better.

The best AMD system that meets the three requirements above was the one I cited. It's not fair to compare a 2 JVM system to a 1 JVM system.


Just a brief note about what I think is fair for comparisons. First of all, in an ideal world, everything would be the same except the hardware. That never happens though. So there are a couple of options then:

1. Compare best configurations
2. Compare the most similar system
3. Some combination of the two

Now, if you are not doing the comparison and measurement yourself and just want to compare to what someone else has done, then using the best external configuration starts to make more sense, since the probability of finding something similar is slim.

DK