All,

Thanks very much for many good advice to help speed up my sorting 10 million record issue. I tried a few alternative and found a most reliable way to reduce the time, in my case, is to cut the large dataset to a few small datasets and then merge them back. And I'll explain why I call it "reliable way" below:

As you may imagine, I use the dataset for many other analysis. One puzzling observation is why the SAS performance in terms of processing time can vary so much. For a simple procedure, say, a index creation, or regression by a group variable, the time can vary from about 40 minutes, to 12 hours! I understand if the speed can be affected by factors such as how much other work I use the computer for, say, checking emails in the same time, etc, but the variation is too large to be explained by these contributing factors.

Does anyone have some insights why this is so? Unlike a sort problem I can always try to find a better way to do it, this variation issue seems harder for me to deal with, since a seemingly perfect code can take magnitude longer simply I use it at a different time.

Again, thanks!

BTW, if it doesn't cause you a lot of trouble, I'd prefer any of you who can share your wisdom to cc to my email address. It's not very convenient for me to go to SAS archive to extract related reply, besides, I'd save them for later use.

A.

__________________________________ Do you Yahoo!? SBC Yahoo! DSL - Now only $29.95 per month! http://sbc.yahoo.com