On Thu, 8 Jul 2004, Stas Kolenikov wrote:
> Dear StataFolks,
>
>
> My first idea was to do something about the temporary files. I don't know
> enthusiastic about it.) The best option of course would be to have some
> sort of RAM drive, so that -preserve- would mean copying a segment of RAM
> to RAM rather than to hard drive. My guess would be that about half of
Is this a class lab or a research computer? Either way I would be
surprised if this worked out well. In a research computer, the RAM disk
will cut the maximum size of a dataset in half (RAM would have to be big
enough for two copies, one in stata and one in /temp) which will cause
problems as many really big datasets exist. In a classroom lab, the /temp
would need to be many times the size of the largest dataset, since several
might be saved at once. But that presumes that the analyzed datasets are
quite small, or preserves will start to fail for lack of disk space. Any
time you overflow real memory into virtual memory or swap, runtimes
increase by 3 orders of magnitude, so you don't want that to happen very
often.
In our experience "preserves" are not a major source of time use.
Virtually all long running jobs use 100% of the CPU, and short running
jobs don't really matter that much. A teaching environment might be
different.
I can offer the following observations of Stata from our experience with a
variety of hardware in a research environment:
0) The largest dataset you can analyze is about 65% of real memory
under unix (and lots of people will have datasets larger than that).
1) Performance is proportional to clock speed, for Pentium or Sparc.
The Sparc is not superior per cycle.
2) Dual processor machines can do 2 jobs with very little loss of speed
3) Hyper-threaded Pentium CPUs can do 2 jobs with slight loss of speed.
The last point is very surprising to me.
Daniel Feenberg
*
* 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/
