MathGroup Archive 2011

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Wolfram, meet Stefan and Boltzmann

  • To: mathgroup at smc.vnet.net
  • Subject: [mg117430] Re: Wolfram, meet Stefan and Boltzmann
  • From: Uayeb <uayebswinburne at gmail.com>
  • Date: Fri, 18 Mar 2011 06:00:51 -0500 (EST)
  • References: <ili6f5$kk2$1@smc.vnet.net> <ilksdj$6b2$1@smc.vnet.net>

Considering Daniel's comments about needing to load som .mx files, and
some of my own experience with Mathematica and other packages, I'll
hazard a guess that this is because of Mac OS X's disk caching.

As an example, I use IDL a fair bit (not particularly because I like
it, but because a lot of astronomy code is already in IDL). In the
process of reducing astronomical datasets, I often call IDL from the
command line over and over to do various steps of the reduction. Each
time, IDL is loaded from scratch (terribly inefficient, I know, but
faster than me rewriting my code). The first time it is called, it
often takes some seconds to load, however, all the subsequent
invocations are much faster (under 1 second).

I have always assumed this is because of Mac OS X's disk caching
mechanism. When a file is requested from disk, the contents of the
file are loaded into free ram so that any subsequent request for the
file is handled from ram rather than from disk. (This ram usage is
labeled "Inactive" in activity monitor, from my understanding, and is
dumped as soon as an application requests more ram).

A speed consious user will also note that for a file being regularly
accessed, putting the file on a "Ram disk" often has little, if any,
speed improvement than leaving it on a physical disk, except for the
first access.

My guess, therefore, is that most of us using Mathematica on a regular
basis, will not see the associated speed delay AES is mentioning even
if we quit Mathematica because many of the relevant files are still
cached in ram. In previous (non-Mathematica=96Word loading times,
actually) discussions of application performance, my understanding is
that the only way to eliminate the OS caching impacting your results
is to run the test on a freshly restarted machine.

Unless I've missed it, AES has not commented on whether subsequent
evaluations of the integral are any faster, or if this happens
everytime s/he restarts the kernal, which might be more interesting.
Afterall, much of the performance of a modern computer is wrapped up
in all the various levels of caching, mostly because much of what we
want to do is the same or very similar to what we have just done. So a
purely unchaced senario is not particularly representative.

Cheers,
Andy

On Mar 16, 10:39 pm, AES <sieg... at stanford.edu> wrote:
> In article <ilnh9v$ob... at smc.vnet.net>, Roman <rschm... at gmail.com>
> wrote:
>
> > AES,
> > mine is taking 18 seconds as well, and I have a MacBook as well. Maybe
> > a Mac hardware thing?
>
> Thanks for confirmation.
>
> For the record, I'm absolutely not pushing any agenda here -- just
> curious.  Macs and Mathematica do many things so blazingly fast it's
> near unbelievable.  And this seems like a relatively simple, smooth
> integral using only simple, smooth, commonplace functions.  So why does
> evaluating it take so long?



  • Prev by Date: Re: Changing default options of functions in a package when loading it
  • Next by Date: Re: Does MemoryConstrained count stack space?
  • Previous by thread: Re: Wolfram, meet Stefan and Boltzmann
  • Next by thread: Re: Wolfram, meet Stefan and Boltzmann