Consider creating an index for the dataset. Proc Sort moves entire records, which can be time consuming when the record count or record size is large. An index creates a parallel file with only the sort key and a pointer to the original record; that's much less data to sort. After the indexing, DATA steps would use a BY statement to read the records in the sorted sequence.

I recently inherited a SAS program that sorted a file 15 times; there were 100k+ records and 150-175 fields in the record. I replaced the first sort with a Proc Datasets/Index and removed the other sorts. That single change reduced runtime from over an hour to several minutes.

Hope this helps.

--Andy Arnold

---- "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM> wrote: > On Mon, 19 Nov 2007 11:00:25 -0500, Wensui Liu <liuwensui@GMAIL.COM> wrote: > > >the way i am doing is > >proc sort data = xxx sortsize = max;..... > > > >but i am not very happy with the performance with large data. is there > >a way to do a faster sort than proc sort? > > > >appreciate your insight! > > People here have often recommended Syncsort.