Date:         Fri, 26 Apr 1996 13:58:00 -0700
Reply-To:     JAMES@UCRAC1.UCR.EDU
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         "S., James" <JAMES@UCRAC1.UCR.EDU>
Subject:      Re: Windows NT Optimization
Comments: To: KLASSEN@UVVM.UVic.CA

>
>>From:         Jack Williamson <104316.2173@COMPUSERVE.COM>
>>Subject:      Windows NT Optimization
>>To:           Multiple recipients of list SAS-L <SAS-L@uga.cc.uga.edu
>>
>>Does anyone know if there is anything one can do to improve the
>>performance of SAS 6.11 under Windows NT?
>>I have a regression with 35,000 obervations and 40 variables that
>>takes 3 hrs to run on a pentium pro 150mhz, w/80meg RAM, 2gig
>>SCSI HD. I must be doing something wrong.  Thanks.
>
>
Just a wild guess (given that most folks have reported sub-1-minute
completion times), is it possible that some  of the memory chips are
faulty?
 
OS/2 is very finicky about memory: it can malfunction if you so much as mix
chips of different speeds, e.g. 60ns and 70ns. It has also been reported to
barf at memory chips from different manufacturers. Sometimes a faulty hard
disk controller card has been know to prevent OS/2 from addressing memory
above 16Mb.
 
Since both NT & OS/2 are 32-bit operating systems, I wonder whether both
are equally intolerant of flaky hardware.
 
Also one could embellish the fragment of code I posted earlier by (1)
wrapping the whole thing in a DO-loop -- in a macro and have it run, say 5
times; and (2) sticking a SELECTION=RSQUARE option in the model statement.
 
From (1) you'll get average completion times, after eliminting the first
iteration, as it'll suffer from the overhead -- albeit tiny -- of loading DLLs
for the first time.
 
james ssemakula
TeamOS2
=======