Dear SAS-L:

I was intrigued by some recent postings on sorting. Looks like in specific cases a self-coded routine can run faster that PROC SORT. I am quite interested because daily we receive hundreds of unsorted vendor files in the form of SAS datasets. The files are then used as drivers in further processing, so we sort them with NODUPKEY and eliminate out of range keys in the WHERE clause. It takes lots of resources even though the datasets contain 1 variable, ID, which is just a positive integer. In all, we have about 350,000 accounts (this can probably grow to about 500,000 due to acquisition). IDs from 1 to some MINID (1000 but may vary) are reserved, so the WHERE includes only keys between MINID and whatever MAXID is at present (350,000 now but may grow to 600,000 due to acquisition). Out of 200,000 to 300,000 observations, in average, 25-50% end up being deduped. If we could cut CPU time even 10-15% coding a custom routine would be worth it. We run SAS609 on mainframe.

Any fruitful ideas? Maybe somebody has already done something like that?

Al