[aarch64-port-dev ] JTREG, SPECjbb2013 and Hadoop/Terasort results for OpenJDK 8 on AArch64

[Edward Nevill]

On Wed, 2014-06-25 at 10:32 +0100, Andrew Haley wrote:
> On 06/25/2014 09:21 AM, Edward Nevill wrote:
> > Relative performance: Server critical-jOPS (nc): 1.33x
> 
> Umm, is this real?

Probably not.

There is a large variance on the SPECjbb2013 results. The critical jOPS has varied between 0.90 and 1.33, max-jOPS shows less variation between 0.93 and 1.03.

This is to be expected because critical-jOPs can be negatively influenced by a single event (eg a GC event), whereas max-jOPS is less influenced.

It might be worthwhile trying G1GC which should have less jank. Also possibly looking at -Xmx parameters to see if we can get more stable results.

All the best,
Ed.