[Sbcl-help] cl-bench Windows results SBCL vs. Clozure

[Brandon V. E. <bva...@gm...>]

On Fri, Apr 15, 2011 at 12:13 PM, Roman Marynchak
<rom...@gm...> wrote:
>
> So, how fast is SBCL on Windows? I believe that the performance is decent.

I can't run report.lisp on any of my Windows CLs, so unfortunately at
present I can only provide raw output.  I've attached both SBCL and
Clozure CL, the most interesting cases.  Note the .txt extension to
keep GMail happy; if you try to analyze these with report.lisp, remove
the extension.  My machine is a 2GHz Intel Core 2 Duo laptop with 2GB
RAM running Windows Vista.  I don't know what these numbers mean in
absolute terms, like compared to another machine, or to another
language like C.  I can say that SBCL is faster than Clozure in some
areas, and slower in others.  Clozure is way slower at the BIGNUM and
PI-DECIMAL benchmarks.  SBCL is somewhat slow for STRING-CONCAT, and
as we established earlier, piggish in heap footprint.  Clozure does
better on most of the CLOS tests.

I'm not seeing a "slam dunk" to pick one implementation over the other
for performance reasons.  My experience so far is that Clozure has the
better bug-freeness on Windows.  They fully support the platform,
whereas SBCL is in a state of "permanent ongoing port" to Windows.

A caveat for Windows Vista benchmarking: the system has a nasty habit
of checkpointing to the hard drive in the background for long periods
of time.  Fortunately on my laptop this is quite audible, so I don't
think I ran the benchmarks while this was happening.  Just be aware
that it can happen, that background events on Vista can randomly ruin
things in a big way.


Cheers,
Brandon Van Every