On Wed, Nov 12, 2008 at 10:24 AM, Wenlong Li <[EMAIL PROTECTED]> wrote: > btw, I tried your delayed recompilation patch, and found it doesn't > bring performance benefit for my test case. That was Alexey Fedotov, but the patch is mine ;)
> After checking the code, > it sounds like the recompilation doesn't happen as the profiling > result doesn't exceed the threshold. So recompilation is not time > consuming in VM creation. Once again, what's consuming most of the time on executing SPECjvm2008:startup? Is there methods from workload that ought to recompile but does not? What's the compilation overhead? Without that data, all we saying is speculation on general case :) Thanks, Aleksey.
