Hardware suggestion

[Cuneyt]

I am looking for hardware configuration suggestions to build a workstation that will run SAS at the highest speed possible.

I am currently running base SAS and SAS/STAT on a desktop running Windows 7 64-bit. However, as my data sets grow, the processing times have become unacceptable ( > 30 min ). As an example, I am running PROC HPMIXED with a data set with about 100,000 observations and about 30,000 subjects (variables). I anticipate even larger data sets in the future. I do no graphical/data mining/big data/text-processing work. I don't have large storate needs. A 1 TB hard disk should be plenty for my need. I know little about hardware. Here are my questions:

1) I think a workstation with a lot of processing (CPU) power and RAM would greatly reduce the processing times. Does this make sense?

2) On the Internet, I have seen 12-core CPU's and motherboards that can support two CPU's. Would SAS be able to take advantage of a workstation with two 12-core CPU's?

3) In terms of memory, I know that 64-bit can address over 100 GB of RAM. Could SAS potentially use all of this RAM?

4) Would I need any GPU's, given that I don't do any graphics work?

5) To keep things simpler, I want to stick with Windows and avoid any kind of distributed processing.

Any suggestions are greatly appreciated.

Sincerely,

Cuneyt

[boemskats]

Hi Cuneyt,

If you are building a workstation for yourself and have the budget, I'd suggest the following kind of thing looking at your usage:

• An LSI Megaraid SAS card, something like the 9271-8i. Maybe two of them to spread the load and if you really want dual CPUs
• 4-6 Samsung SM843T or SM863 SSDs to plug into it. As you double capacity you double the number of write cycles and double MTBF. I know you say you don't need much storage but theoretically if you can afford bigger it can be better in terms of longevity because of the number of times it will get overwritten (as SASWORK). For performance though, and if you're on a budget,  having a greater number of smaller disks is better than having a few big ones. Configure them in raid on the controllers - raid0 for work/util and raid1 for your persistent storage. 
• With the above you should get otherwordly performance with a modern consumer processor like the i7 6700k, unless you plan to run loads of jobs in parallel. If that's the case and your budget is unrestrained you could get a dual socket Xeon board and more cores, but make sure you get 2 LSI controllers and 8-12 SSDs. Licensing cost can also affect your budget here.
• However much RAM you use, however much RAM you can afford. What do you set your MEMSIZE to at the moment? SAS can use loads of RAM, and with Windows SAS you can also use MEMLIB so that can speed things up nicely. But it's quite expensive. I'd guess you should be fine with 32GB ish.

I haven't profiled PROC HPMIXED yet and don't know if it's multithreaded or not, but might do tomorrow just to see if the above advice is way off. This is more of a general purpose workstation spec, which to be honest I personally think is overkill, but if you can afford it (which it seems you can) I think you'll be very satisfied with. It's extremely impressive what 8 threads and a few SSDs can do. 

Windows is fine. You don't need a GPU.

Hope this helps.

Nik

[Cuneyt]

Nik,

Thanks for all these suggestions.

Cuneyt

[Reeza]

Your complexity seems to come from the subject level, not the number of records, 100,000 shouldn't be much. 

Given that I'd suggest highly increasing your RAM becuase I'm fairly certain those types of processing are very RAM dependent.  

I run a 16GB and use a 500 GB external SSD, i5 quad core processor and that's been fantastic for speed.  

[Cuneyt]

Reeza, thanks for the helpful suggestion. I have a similar processor like you, but only 8 GB of RAM. I'll try to increase the RAM size.