Comparing platforms

[Sven N. Thommesen]

Hi all!

In the process of updating the random library a tiny bit, we've had
occasion to run some timing tests on the random library objects on various
platforms. I report some of the results here for everyone's entertainment ;-)

All timings below are in microseconds per method call, so smaller is better.
(Timings approximate only, since other processes may have stolen cpu time.)

                         PPro200     PPro200     486-66      486-66
                         Linux       Linux       Linux       Linux
                         (none)      -O3         (none)      -O3

MT19937 (Unsigned)        0.6248      0.4049      4.0737      2.8580
MT19937 (thinDouble)      0.9433      0.6700      5.9959      4.4981
MT19937 (longDouble)      1.7966      1.1760     11.0074      8.2753
RandomBitDist             0.8917      0.6093      6.0610      4.7801
UniformUnsignedDist       1.2069      0.8017      7.5652      5.3587
NormalDist                4.5146      2.9832     26.8641     20.7184
GammaDist                20.5054     14.4654    123.5773     96.8614


                         Alpha       Alpha       IndyIP22    UltraSparc1
UltraSparc2
                         NetBSD      NetBSD      Irix        Solaris
Solaris
                         (none)      -O3         -O3?        -O3 -mv8
-O3 -mv8

MT19937 (Unsigned)        2.6202      1.3098      3.0426      0.8876
0.5865
MT19937 (thinDouble)      3.5646      2.3440      4.2347      1.2444
0.8339
MT19937 (longDouble)      6.8918      3.2083      7.5072     44.9798
30.6577
RandomBitDist             5.3116      1.6797      4.3328      1.4739
0.8514
UniformUnsignedDist       7.2316      5.3633      5.2007      1.8591
1.1647
NormalDist               22.1530     16.5018     16.9574     72.1065
44.2907
GammaDist                93.9495     70.5389     76.0339    326.2035
212.7963

Some obvious comments:

a) optimization (-O3) matters!
b) A Pentium Pro is not a bad place to be -- 
   and I expect a 300 MHz P-II will really smoke!
c) Sparcs seem to have trouble with long doubles ...

Cheers,
Sven Thommesen
Auburn University


                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.
                  ==================================