Team PLClub ICFP entry -- comparing the performance of OCAML and SML

Lal George george@research.bell-labs.com
Thu, 12 Oct 2000 21:42:42 -0400 (EDT)


sweeks@intertrust.com(Stephen Weeks) wrote:
 > ....
 > 
 >                       my original    with jhr
 > 		      translation   modifications
 > 			    110.29         110.29
 >               OCAML  MLton  SML/NJ  MLton  SML/NJ
 > holes           1.8    3.5     5.0    3.2     3.9
 > fov             1.5    2.1     4.4    1.8     3.2
 > intercyl        1.6    2.4     6.0    2.1     4.3
 > snowgoon        2.9    4.0     8.4    3.3     5.1
 > dice            3.9    5.7    10.8    4.9     8.8
 > golf            1.5    2.5     4.2    2.4     3.1
 > cone-fractal    3.7    4.9     8.4    4.3     6.5
 > large           4.3    3.1     7.6    3.0     6.7
 > pipe            5.4    5.3    11.3    4.6     7.9
 > chess          16.0   17.8    38.6   15.5    21.6
 > fractal        12.2    8.7    41.1    8.5    45.4
 > 
 > geom-mean       3.6    4.4     9.5    4.0     7.1
 > 
 > ....
 >
 > I reran the fractal and large benchmarks for the original translation under
 > SML/NJ 110.25 and got roughly the same results.  Here are the numbers.
 > 	
 >          110.25
 >          SML/NJ
 > large       3.9
 > fractal    26.0

I am confused. Perhaps I missed a message inbetween. 

You say you "got roughly the same results" for large and fractal,
however the numbers from 110.25 are nearly twice as fast as those for 
110.29. Which set of number (those in 110.29 or those in 110.25) are
the correct ones?

In some sense the question is mute, as SML/NJ is almost universally
two times slower than CAML.