[Mono-list] Re: The Mauve unicode testcase and VM performance

Ben Maurer bmaurer@users.sourceforge.net
Tue, 13 Apr 2004 22:41:29 -0400

Ok, there are a few factors to consider here:

JITing takes a hit *once* -- remember, once a method is jitted, there is
0 cost. For something like SciMark, you are running the test for a
pretty long time. So, if there is a .01 second delay in jitting it on a
1 second test, you wont see anything.

AOT can cause code to be aligned differently, because it emits some
slightly different constructs. This can give misleading results.

There are some cases where the AOT emits code that we could work on
optimizing. However, this is mostly not in the areas that SciMark tests.

Anyways, having numbers is very valuable. I would love if you were to
post your benchmark results from week to week.

-- Ben

On Tue, 2004-04-13 at 22:24, Marcus wrote:
> I get almost identical results overall on SciMark with AOT and JIT 
> compilation. The overall results are 97.0 MFlops (JIT) versus 95.4 MFlops 
> (AOT). Some subtests are faster with JIT; some are faster with AOT. This 
> difference is less than what I see from week to week comparing JIT-to-JIT, as 
> various components of the runtime change.
> On Tuesday 13 April 2004 5:08 pm, Jonathan Pryor wrote:
> > AOT (or pre-JITing) assemblies is not as useful as you'd think.  It has
> > two primary effects:
> >
> >   - Reduce memory requirements, as memory doesn't need to be allocated
> >     for the JITed code
> >   - Decrease startup time, as (again) the JITing doesn't need to be
> >     done.
> _______________________________________________
> Mono-list maillist  -  Mono-list@lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-list