- Comparing SAS installations.
Define standard ways of testing.
SAS base code
(Phil Rack) This is response to a thread on another forum that I responded to months ago. I think it's relevant here so I'm basically re-posting it.
It seems that it's common to ask the question, "What do I need to do to improve my SAS processing performance?" Or, as the thread you started "How would you stress test a SAS Server?" They are good questions to ask considering how much statistical processing software can cost today.
A while back, I started to write a SAS language program that I could take to a client site and see rather quickly what I could expect from the hardware I was using. It's nothing fancy but it does give a glimpse into what you can expect from a workstation or server in terms of processing data.
One thing I would like to see is some standard program a person could run to try to get an idea of how their hardware and environment stacked up. There are all kinds of programs out there such as Prime95 that kind of torture tests your CPU. But the SAS Language is different in that we have more to worry about than CPU utilization. Disk I/O is often a hindrance to efficient processing.
The program I wrote will run on WPS 3.0x under Windows, Linux (tested on Redhat, Fedora, Ubuntu) and Mac OS X. (I know that SAS doesn't support OS X natively but WPS does.) I believe it will run properly on SAS on Windows. At least it did the last time I tested it on SAS. I don't have a SAS System to test this on at the moment. I would suspect that it would also run on other OS platforms as well but again, I don't have access to SUN Solaris or AIX, etc... A program such as a Work Station Assessment could be valuable to users to identify where there are bottle necks and also folks on the different list serves and forums who could comment on the results of such a program in recommending where improvement can take place.
Here's a link to a copy of the output from the program.
The actual program: http://minequest.com/downloads/performanceassessment_2012.txt
The idea was to try to test just the most basic aspects of the systems. I never know what the client or user may have licensed so I kept the assessment to what could be run in the BASE language. I wanted to try to see if the system could multi-thread efficiently and test I/O throughput.
As written, I think letting the program go out into the public domain could be useful, but I would also like to see it further refined. If anyone is interested in refining it, I would be happy to test it and make some supportive comments.