[Mono-list] Re: tiny benchmark of current mono clr, java clr, ms clr

[Francesco Delfino]

I think that if you just change the Console write operation with a "swap"
operation like this:
    tmp = a;
    a = b
    b = tmp;
you can have completely different results. And even with this test, you just
are testing how mono o .net or java deals with very few of their instruction
set.
Besides your admitting that the tests are without any significance for every
application but your, your application benchmark (testing the console output
performance) is useless for most other different applications and may lead
to confusion.
We at Tipic, Inc. are trying to evaluate Mono Performance on a real
application (at designing/alpha stage) and we got that actually both MS libs
and MS.Jit ouperform Mono on small/simple benchmarks; but the performance
gap gets smaller when some "real" (complex) algorithms are done (maybe
algorithms different from what is already implemented in the CLR ;-) ): this
make us think that Mono is designed to be a general purpose Jit, optimizing
everything with the same accuracy, while .NET, instead, does some operations
with much more accuracy.

--
Regards,
   Francesco Delfino
   Tipic Inc.
   http://www.tipic.com
   e-mail & Jabber: pluto@tipic.com
   blog: http://www.tipic.com/blog/pluto@tipic.com




"Daniel Mettler" <mettlerd@icu.unizh.ch> wrote in message
news:Pine.LNX.4.30.0211102119160.5463-100000@banana.icu.unizh.ch...
> re all,
>
> On Sun, 10 Nov 2002, Guenther Roith wrote:
> > You shouldn't print anything to the console, because this tests how fast
> > your os can scroll the terninal window.
> > Do somehthing else like only counting.
>
> well, it's certainly affected by, but obviously this tiny bench
> does not solely measure the performance of the terminal/console
> and its scrolling. else the measured differences (most of them
> being significant at a confidence level of 95%) between the
> runtime environments of the same os platform could not be
> explained.
>
> the bench could be much better of course :) admittedly, i did not
> take the time to write a nice, more or less representative
> benchmark (which is a difficult "political" task btw). but this
> was not the goal anyway. i just wanted to quickly compare the
> relative performance of the runtime environments for that
> particular task. there is no claim for representation or
> generalization at all. and i did not want to compare linux vs.
> windows either (perhaps the second point in the conclusions is
> misleading, i think i should remove it).
>
> nevertheless i was surprised by some of the results. regardless
> of the influence of the terminal performance itself i expected a
> "home run" of the ms clr on windows. but mono appears to be
> significantly faster (in the context of this specific task). as
> well, i did not expect mono to beat java on windows (dito). i do
> not know the reason for this however (perhaps you know?).
>
> another thing which surprised me were the quite different cil
> file sizes generated with the mono and ms compiler respectively.
> i did not further analyze this though.
>
> regards
>
> dan
>
> --
>       ...::: Daniel Mettler | http://www.numlock.ch :::....
>
>
>
> _______________________________________________
> Mono-list maillist  -  Mono-list@ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-list
>


---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.408 / Virus Database: 230 - Release Date: 24/10/2002