SAS Export As Step - Performance

[BlayLay]

Hi All,

Quick question: is it normal to see very slow performance when utilizing ‘export as a step’ functionality, when exporting a 4M-row 20-column dataset to a mapped network drive? I’m talking several hours…I’ve read on other posts that saving locally versus on the SAS server, although this network drive is more “local”. I’ve also seen others suggest PROC EXPORT. Is there anything I can do to improve performance, or is this simply the reality with such a large dataset?

I’m back to SAS EG 8.3 although I’m dreading the SAS Viya migration…

[ballardw]

What does the log show, including the code?

"Normal" may depend on your network configuration and/or traffic load and possibly space available on the network drive. I would say that sounds excessive though.

I would discuss this with your IT folks to see if they know there are any issues with performance on that drive.

I used to make copies of shared data sets on my local drive that only had 4500 observations though a couple hundred variables because the network as configured was just plain slow and reading those for something like Proc Freq could take 15 or more minutes. Since I was using those sets literally 100's of procedure calls daily that performance was unacceptable. IT eventually upgraded the drives and network and the performance became usable.

[SASKiwi]

Are you by any chance exporting as a CSV and using the GUESSINGROWS option? This option is SLOW if done on the lot of data. Check your SAS log to confirm if this is being used. It would also be a good idea to post the log of this step in your post so we get a better idea what is happening.

It would also be a good idea to discuss this problem with your SAS administrator who would have a better idea of your network performance.

[Kurt_Bremser · Accepted solution]

At the moment you have two concurrent network streams competing for bandwidth: the download from your SAS server to EG, and the output from EG to the network share. If the share is also available on the SAS server, writing to it directly with PROC EXPORT (or a DATA step you write yourself) will perform better.