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

st: mata speed on SUSE linux64


From   palangkaraya-half <alfonsebay@yahoo.com.au>
To   statalist@hsphsun2.harvard.edu
Subject   st: mata speed on SUSE linux64
Date   Mon, 06 Jun 2005 16:34:29 +1000

Dear Stata users,

I did two kind of benchmarks:

(1) a "synthetic" one using stata's new matrix language, and
(2) a "practical" one consisting of only data management tasks.

The first benchmark randomly draws Y vector and X matrix from a uniform
distribution, and then uses the QR decomposition to solve for the
generalised least square of Y on X.

Four tests were conducted depending on the size of X and Y and
the number of repetitions.

Test 1: Repetitions=  10, Y= [1000 x 1], X=[1000x1000]
Test 2: Repetitions= 100, Y= [1000 x 1], X=[1000x1000]
Test 3: Repetitions= 100, Y= [ 100 x 1], X=[ 100x 100]
Test 3: Repetitions=1000, Y= [ 100 x 1], X=[ 100x 100]

The "synthetic" results are:

PC (CPU, Speed, Memory, OS, Stata Version)           Time (minutes:seconds)
                                                     Test1 Test2 Test3 Test4
---------------------------------------------------------------------------
Athlon 64 FX/55, 2.6GHz, 3.0GB, Windows X64, SE/9/64, 0:18, 3:09, 0:0,   0:3
Athlon 64 FX/55, 2.6GHz, 3.0GB, Windows XP,  SE/9/32, 0:21, 3:27, 0:0,   0:4
Pentium 4,       3.0GHz, 2.0GB, Windows XP,  SE/9/32, 0:24, 3:57, 0:1,   0:4
Athlon 64 FX/55, 2.6GHz, 3.0GB, SUSE Linux64,SE/9/64, 0:45, 7:45, 0:0,   0:6

The second benchmark manipulates a data set of 14811 observations (6.5MB in
memory usage). This is actually an unbalanced panel data set of six years
and approximately 9000 firms. The main goal of the benchmark is to count the

number of neighboors of each firm which is located in the same suburb, the
same city, and then, the same state. Along the way summaries (such as
average
size) of the neighboors were construction. Because of the complexity of
the measures constructed, the do file needs to iterate one firm-year at
a time.

The results are:

PC (CPU, Speed, Memory, OS, Stata Version)           Time needed to complete
                                                     (hours:minutes)
---------------------------------------------------------------------------
Athlon 64 FX/55, 2.6GHz, 3.0GB, SUSE Linux64,SE/9/64,      1:07
Athlon 64 FX/55, 2.6GHz, 3.0GB, Windows XP,  SE/9/32,      1:16
Athlon 64 FX/55, 2.6GHz, 3.0GB, Windows X64, SE/9/64,      1:15
Pentium 4,       3.0GHz, 2.0GB, Windows XP,  SE/9/32,      1:49
Pentium 4,       3.0GHz, 2.0GB, Windows XP,  SE/8/32,      1:51

I have attached the files to do the synthetic benchmark if you are
interested.
You will need stata 9 and the additional "elapse.ado" installed
(enter this into your stata command if you are not sure:
"net install ip24_1.pkg").

Then, run "looptest.do" followed by "benchmark.do"
If you do these, please let me know your results and your PC information.
Also, if anyone knows any idea why my linux results are very slow (even
when run in a batch mode)

Thanks.

-alfons

ps: i did not attached the do files for the second benchmark due to
data use restrictions.

Dr. Alfons Palangkaraya
MIAESR, Univ of Melbourne
Ph.  +61 3 8344 2119
Fax. +61 3 8344 2111
alfonsp@unimelb.edu.au
http://www.melbourneinstitute.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