Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: linux, windows & informal benchmarks


From   ariley@stata.com (Alan Riley)
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: linux, windows & informal benchmarks
Date   Tue, 10 May 2005 10:03:18 -0500

Scott de Marchi (demarchi@duke.edu) asked about the relative performance
of Stata under 64-bit Windows and 64-bit Linux running on an x86-64
platform:
> My prior was to use linux with Stata 9, but I did some very informal
> benchmarking before leaping (n.b.: I have both a linux 64-bit and a
> windows x64 license).  I ran comparisons using the same dual boot machine.
> One drive had windows server 2003 64-bit and the other had the 64-bit
> version of ubuntu.  Hardware: tyan k8we, 2x252 opterons, 8x1gb ocz server
> ram, and 1 74gb sata raptor for each os config.
> 
> On a suite of operations that were mle regressions, histograms, summaries,
> and some user written code to generate comparisons to benford's law for
> large data sets, I found that the windows version of Stata 9 did much
> better overall than linux (over the suite, linux took over 50% longer to
> finish).  I haven't broken this down by task or thought real hard about
> weightings of the various items in the suite.

It is difficult to comment more on this without knowing more about the
tests that were run, but we are very surprised to hear that Linux took
over 50% longer to finish.

I would like to correspond privately with Scott to experiment on his
system and perform more timing tests to determine if there is something
causing such a speed difference on his system.

Some initial questions that come to mind are

   - did Scott use the console or the X Windows version of Stata under
     Linux?

   - did Scott draw a lot of graphs?  If so, was the graph window
     bigger under one operating system?

   - did the tests generate a lot of output in the Results window?
     If so, was the Results window bigger under one operating system?
     What was the screen resolution, color bit depth, and video card
     driver used under each operating system?

   - were some ML routines run that perhaps took a different path
     to convergence on each system due to different math libraries?

   - how close was each system to the limit of memory on the computer?

If Scott will email me, perhaps we can do some tests on his system
this week and report back to the list on any discoveries we might
make.


--Alan
(ariley@stata.com)
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index