| This post was kindly contributed by The SAS Training Post - go there to comment and to read the full post. |
Welcome to 2011! A new year, and for me, a new blog. My name is Michelle Buchecker and I have been using SAS since 1988. No, I’m not old, I can’t be. There must be just some weird time warp involved.
One of the aspects I love about my job is the out of the blue questions that customers send to me. Most of them have to do with previous SUGI/SAS Global Forum papers I’ve written. I’d like to share one with you, and my response.
Question: I read your paper on Pipeline Parallelism. I was wondering if you could provide an example of using pipeline in batch mode on a platform like AIX where rsubmits aren’t used. I’ve fiddled with it some and I’m not having much luck so I decided to stop beating my head against the wall and ask for help.
Answer: Pipeline Parallelism always requires an RSUBMIT even if it is on the same physical machine. Think of it as your original SAS session sending code to a separate SAS session on that machine. The code in that paper actually is the code that you would use on a single machine with multiple CPUs. You may want to change:
to:
as that invokes the child SAS sessions using the same executable/shell script as the parent session.
You may note that SUGI 30 was in 2005. But due to that weird time warp again, several SAS users are still not aware that they can have a DATA step and the PROC SORT on the data going at the same time without writing the data to disk! This technique saves both disk space and CPU time. Hmm, saving CPU time??? Maybe that’s where the time warp is going!
| This post was kindly contributed by The SAS Training Post - go there to comment and to read the full post. |