Which Procedures to test PC speeds

[deleted_user]

Hello, in my team at work we all have a variety of computers of different ages and specs. We run all SAS code locally and I would like to test the performance of each of the machines. To try to identify which computers are better for certain things.

Could you tell me which procedures are best for testing;

I will probably use proc sql to extract a large dataset across the network and store this locally.

I then want to do something to test the cpu, ram and i/o for the disks.

What would you suggest? Plus anything else I could probably do, I know there is an option to produce some more detailed stats on run time. Not just cpu and real time.

Cheers.

[Doc_Duke]

On a PC, there are a lot of other things that effect performance than the SAS procedure itself. Some thoughts.

PC performance degrades with time. Microsoft and other applications deposit things on the PC that effect performance. For a level playing field, the PC's should be re-imaged.

McAfee On Access Scan sucks resources. It is particularly noticeable when first running SAS. McAfee checks all those little DLLs before SAS can first use them. Again, for a level playing field, either turn off the OAS for the SAS Program Files and data sets or re-boot between tests.

Defrag all the disk drives before starting the tests.

[deleted_user]

No offence, but that's IT's job. As a user I want to say to IT, "this is the situation and it's not good enough, these are the things I need more of."

If doing the things you suggested does also provide an overall increase in performance then that we can do every six months.

But in the mean time I would like to identify bottle necks within the system itself. for example if my RAM is below par compared with the rest of the system then this can be increased.

Also because they are a variety of different machines, I can then figure out what the best configuration is and potentially have my IT build some new machines.

Regards

[SASKiwi]

I suggest you have a look at the test programs SAS provides to test that a SAS install has worked correctly. There is a different program for each product so the nice thing about these is the broad range of tasks tested so they should test all aspects of your PCs performance. A search through the SAS install directory for SAS program names including the word "TEST" should locate these.

[deleted_user]

I've had a look at them, but they are all very quick and the differences between will be very small, so will not show up any real differences in performance.

Is there a paper that simple states what each procedure uses in terms of resource ie;

proc ram cpu i/o etc...
sort 50% 10% 40%
means 20% 70% 10%
univariate 30% 65% 5%
etc...

This would help as I could then create a test, and based on the results give a score to each machine for the various components.

If that makes sense.

Regards
Rhys.

[sbb]

It's most appropriate that you would have your own "benchmark test" of whatever SAS application is suitable to your needs - yes, SAS is an application environment that servers different customers with their appropriately developed/maintained solutions. I would not want "IT" to be testing my SAS application performance, frankly.

Scott Barry
SBBWorks, Inc.

Benchmark (computing)
http://en.wikipedia.org/wiki/Benchmark_(computing)

[SASKiwi]

If you are simply wanting a quick to produce, generic test of PC performance I would be tempted to use the test programs but repeat them in a loop so that they take enough time to be able to do comparisons between PCs. If you looped these programs so that they took say 30 minutes, then you would start to see significant variations between PCs.

[Peter_C]

rather than exercising the sas syntax compilers, how about some small code with "big" data?
See: Paper 263-2009: Kros, Don
http://support.sas.com/resources/papers/proceedings09/263-2009.pdf
Very Large Dataset Performance as Measured Across Multiple SAS 9.1.3 Data Storage Options.
Here "Very Large" is approx 500M rows.
The results are very interesting.

peterC