Date: Wed, 22 Sep 2010 20:50:59 -0400
Reply-To: oloolo <dynamicpanel@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: oloolo <dynamicpanel@YAHOO.COM>
Subject: Re: Server SAS vs PC SAS, Pros and Cons
Yes, the workstation in our group has an AMD Phenom II X4 965 BE [4 core
single CPU].
the server is a high end Dell unix machine. I don't know the configuration,
but I do know the server costs hundreds of thousands dollars.
In my view point, the major difference between a workstation and high end
server is the storage system. If we really want a workstation that has a
very fast I/O, the cost will be very high, too, may well be above $10K, IMHO.
For example, the killer factor separates this server from a pool of machines
is indeed the storage system, which is just SO FAST that today I benchmarked
it using PROC RANK. I try to rank a data set with 300 real valued variables,
1E6 observations, and the system reported that due to insufficient memory,
utility files are created. Real time used was about 1min 25sec. Now I only
rank 70 variables, in which case memory is just about enough so no utility
files need to be created. The real time used is 18.9sec [CPU time is
~18sec!], that is the time is proportional to the ratio of variables
involved, and the ratio factor is roughly 1. So that when utility files were
created on the file system, it has very little impact on the performance.
Compare to the workstation, it took 3min 36sec to rank the 70 variables, CPU
time was about 40sec only, though. So bottleneck is the I/O.
On Wed, 22 Sep 2010 19:51:59 -0400, Arthur Tabachneck <art297@NETSCAPE.NET>
wrote:
>Liang,
>
>But are you comparing apples with apples? Is the workstation a multi-core
>workstation?
>
>Besides the higher cost for the server license what got me to consider
>going to multiple-core workstations was that so many on SAS-L were
>reporting faster processing than I was achieving .. and they were all
>doing it on workstations.
>
>Art
>---------
>On Wed, 22 Sep 2010 19:18:36 -0400, oloolo <dynamicpanel@YAHOO.COM> wrote:
>
>>a difference I observed, but not documented is that some procedures that
>are
>>not multi-threaded in PC SAS actually can leverage multi cores in a server
>>version.
>>
>>I tried PROC PRINCOMP to conduct a PCA on a moderate size, 1000-by-1000,
>>real matrix. When the commands were executed on our server, the CPU time
>is
>>larger than real time, which means mutli-threading or multiple cores are
>>involved in the analysis; but on our workstation which has PC SAS, the
>real
>>time is larger than the CPU time, which definitly is single threaded.
>>
>>SAS Doc didn't mention PRINCOMP in their list of procedures that can use
>>multi-threading or multi-cores, however. I guess PROC FACTOR may also be
>>improved.
>>
>>Overall, I am so far satisfied with the server environment for analysis,
>>comparing to the PC enviornment in my previous company
>>
>>
>>On Tue, 21 Sep 2010 11:07:39 -0400, Ya Huang <ya.huang@AMYLIN.COM> wrote:
>>
>>>Hi there,
>>>
>>>We are considering to move away from standalone PC SAS to some kind
>>>of server based SAS. We've done some comparison, but I would like to
>>>get more input before we make a final decision. Would appreciate any
>>>comments on why you think server is better or worse, why did you choose
>to
>>>move away or stay with PC SAS. Different prospects are welcome, not
>necessay
>>>technical concerns.
>>>
>>>It would be great if you can also share the basic configuration of
>>>your server and client.
>>>
>>>
>>>Thanks
>>>
>>>Ya
| 
