Date: Tue, 27 Sep 2005 11:54:37 -0400
Reply-To: Larry Bertolini <bertolini.1@OSU.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Larry Bertolini <bertolini.1@OSU.EDU>
Organization: Ohio State University
Subject: Re: Win server config (was Re: recommendations for SAS PC
Content-Type: text/plain; charset=us-ascii; format=flowed
Pardee, Roy wrote:
> [Noob question here--feel free to ignore.]
> I've often wondered whether there was a way (or maybe windows takes care
> of this? Or the drive controller?) to cache blocks of data such that the
> cache would outlive a single sas job.
Maybe using a RAMDISK utility is appropriate.
Not sure if the RAMDISK utilities create "disks" of fixed size, or
if they can expand (to use pagefile) if your SAS datasets grow beyond
> Say I've got 3 different programs that all hit the same bits of the same
> dataset, and I'm going to run them serially. If I SASFILE the dset into
> memory in program 1, I'm assuming that by the time program 2 gets
> running, the chunk of memory devoted to the dset has been freed & is
> gone daddy gone. Now, I can SASFILE it back into memory in program 2 if
> I want, but I'm still re-doing lots of i/o & so the benefits are not as
> good as if there were some way to sort of 'pin' the dset in memory. Is
> there such a way?
> My frame of reference here is the rdbms, which typically has its own
> data cache that it manages (across multiple users even!). I've wondered
> whether I wouldn't be well served to just chuck all my data (I've got a
> very few, enormously huge datasets) in say, mssql, and just sas/access
> it out of there...
If you use pass-through, so that all data manipulation occurs in the
RDBMS, then that's a reasonable approach. But if you've got to pass
the data back and forth, from SAS and the RDBMS (even if the RDBMS is
cacheing the data) I doubt that will be better than storing the SAS
libraries on good old local disk.
I wonder if the SAS scalable performance data engine provides a
> -----Original Message-----
> From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of
> David L Cassell
> Sent: Tuesday, September 20, 2005 6:36 AM
> To: SAS-L@LISTSERV.UGA.EDU
> Subject: Re: recommendations for SAS PC configuration
> BruceBrad@INAME.COM wrote:
>>For most SAS operations I/O is the bottleneck. I would put as much
>>money into the drives as possible. I've had good experiences with IDE
>>RAID 0 (eg a two-disk RAID 0 system almost doubled the throughput
>>speed). I would guess that RAID 5 would be just as fast reading.
>>Sustained transfer speed is the key (rather than seek times). You might
>>get more back for the buck using SATA than SCSI (particularly if the
>>machine won't be in continuous use).
>>When using RAID, I found best performance when I set bufsize and
>>ubufsize at twice the RAID stripe size (see Davis and Ralston
> Bruce makes a very good point here. (As usual!) I/O is usually the
> killer in data manipulations like you want. But there is an additional
> SAS has a SASFILE statement which allows you to load a data set into
> RAM, for much faster access and data mangling. So make sure to cram
> enough RAM into your machine to handle any size of data set you're
> likely to use, and then you can use SASFILE to speed up intermediate
> You still have to read the darn thing once (at a minimum) off your hard
> drives, though. :-)
> David L. Cassell
> mathematical statistician
> Design Pathways
> 3115 NW Norwood Pl.
> Corvallis OR 97330
> Don't just search. Find. Check out the new MSN Search!