SAS and Multiprocessor hardware

[deleted_user]

I have read a bit about Bufno, sortsize and SGIO but I its still not still not clear. What really affects performance in a desktop? Fast HDD (10,000 RPM SCSI) or 4GB RAM or multiprocessors?

Does having dual core or quad cores or dual quad cores in a Windows XP 32 bit OS really make SAS execute faster?

By execute I mean just a simple sort of a large dataset (5-50million obs with 10-30 vars). Are we going to see a dramatic reduction in real and cpu time?

[Doc_Duke]

Check out this SAS note.

http://support.sas.com/techsup/unotes/SN/012/012568.html

If that doesn't satisfy, I know there have been several presentations at the last couple of SUGI's that discussed SAS performance, so you could search them.

If you are still not satisfied, SAS has a performance consulting service that can help on specifics (for a fee, I believe). They were very helpful to us in sizing some new Unix boxes for SAS.

Doc Muhlbaier

[deleted_user]

Thanks for the reply.

When I posted this question it had to do with the under performance of our very expensive workstations (Dell Precision 670) and how perhaps getting an $9K 690 with a total of 8 cores could help or not help with the situation. I had contacted SAS about this issue a few years ago and the only solution they gave me were the sortsize, bufno and SGIO. That marginally helped with the problem.

The real problem was with hardware which took a random SAN engg. who is totally unfamiliar with SAS to figure out. It as as simple as enabling "read cache" on the RAID controller card.

I wish the help desk person had told us about the SAS performance consulting service (we certainly have the resources for that) rather than saying "Well if you have 4 x 10K RPM SCSI drives in RAID 5 and dual-dual cores and 4GB RAM and sortsize=2g, I guess thats as fast as your processes will ever be."

All it took was a H/W guy to tell us what the problem was........after two years of thinking this is the best it could be, coz SAS said so.