Is it possible to force the OS to perform disk read instead of reading a cached copy in memory? I want to know if this can be done in SAS. If the OS is relevant here, then I look for the answer for Windows XP and UNIX SAS.
I am designing a SAS program to process a large SAS dataset(say 3GB). Currently it is very slow so I am working on improving its performance. But after I run the program once, the dataset is cached in the memory. So second time I run it, it becomes very fast. This is really hard for me to tell if the modified code actually is doing better. Got it? Never encountered this problem???
Message was edited by: aha123
There are so many questions and variables that could be impacting performance, it might be better for you to work with Tech Support on this question. They can look at all your code and help you turn on performance monitoring. In addition, they have many different machine configurations on which to try to replicate what you've observed. And they could advise you about techniques that would result in performance gains.
I came to this forum because SAS Tech Support really sucks. Every time it takes at least one day to get back from them for a question. When I questioned them why the response is so slow, then gave me a link for support policy which basically says Tech Support is free so the response time is long.
However, again, I respectfully encourage some determination with a realm of reasonableness with one's query/request for technical support. Come on - "force disk read", expecting a reply, considering various factors like OS, hardware, operating environment, data-conditions, and furthermore with "When the Moon is Blue" and the esoteric-like?
I agree Scott, the more information the higher the chance of getting a positive outcome. But we need to understand that
a) The person sending these questions may not have a solid technical background like you.
b) English may be a second (or third Language)
My comments were just stating that you are paying (many believe it is free and hence have different expectations) for the service, now if you make it hard for yourself not supplying the right information to them then that is another story.
Responses to US customers may be different to the response times in other parts of the world, we have a least a 24hr turn around if our local office has to esculate the question.
I don't share your opinion of tech support
however, I can offer a paper for sas on AIX/Win which demonstrates and measures effect in SAS on AIX and Windows of bypassing the system cache.
(a few years old now, but sounds relevant to your situation) http://support.sas.com/resources/papers/proceedings09/327-2009.pdf
Improving SAS® I/O Throughput by Avoiding the Operating System File Cache
Leigh Ihnen and Mike Jones, SAS Institute Inc., Cary, NC
Beware their conclusions point out that for sas files larger than 2GB you should use SGIO
Shame you were not offered this paper earlier
To the OP: I also strongly disagree with personal experience when contacting SAS Technical Support. For one point, I fully expect that the staff prioritizes an incoming contact/issue/track considering the type and reasonableness of a user's problem and/or question and how it pertains (or doesn't) to the SAS system and its use.
Lastly suggest keeping the forum posts professional and objective, while keeping personal opinions to the self.
I have found SAS Tech support to be quite helpful. Not necessarily the first person, but it ends well. I am a statistical consultant and have often been on the receiving end of calls. May I proffer some advice for better, faster, outcome.
If your question is precise and well focused (unlike this one), then the web interface to tech support works fine. If the question is more of a problem of unknown etiology (like this on), then a call to tech support works better. They can ask the "20 questions" that gets to the underlying problem so they can actually provide a useful answer.
BTW, the problem of testing speed in Windows is not just disk I/O for data, it is also disk I/O for SAS modules, and a plethora of other issues. McAfee, for instance, can really handicap SAS performance the first time around. The only way to reliably test performance related issues in Windows is to reboot between tests (ugly and time consuming, but necessary).
*NIX is even messier for testing. Your performance is dependent on what other people are or have been doing.