IT-DISCUSS Archives

August 2006

IT-DISCUSS@LIST.UVM.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Tim Raymond <[log in to unmask]>
Reply To:
Technology Discussion at UVM <[log in to unmask]>
Date:
Mon, 14 Aug 2006 17:46:37 -0400
Content-Type:
text/plain
Parts/Attachments:
text/plain (70 lines)
I would first check out SAS to make sure it is multithreaded and can take
advantage of the core duo processors.  If it can't you won't see much
performance increase going from 2.4 GHz P4 to 2.4 GHz Core Duo unless the
researcher is running two jobs at once.  The increase in cache will probably
make a difference, the bus speed will help some but probably not much.  Disk
speed won't make any difference unless SAS writes out intermediate results in
very small time increments.

I have some 3.4 GHz core duo Optiplex GX620's that just came in, you are welcome
to load SAS on one of them and try it out if you wish.  They only have 1 GB of
memory but with a reduced problem set you should be able to get a sense of how
they stack up against the original machine.

Michael Kenny wrote:
> Hello
> 
> I have a colleague in my department who is running a series of
> regressions in SAS on a Dell desktop PC. On her somewhat-old 2.4 Ghz
> Pentium 4 these analyses take about 24 hours each to complete. When the
> system had 512 MB of RAM, system diagnostics suggested that the speed
> was limited by a lack of RAM, but now that I have maxed the system out
> at 2 GB, CPU limitations are indicated in the Windows XP Performance
> Monitor. Since this person is about due for a new computer anyway, we
> have decided to look at scientific workstations instead of standard
> desktop systems.
> 
> I am hoping that someone on the list has some experience with
> mathematical analysis on the PC and can provide suggestions on the best
> price/performance ratio for  this type of work. Behind the scenes, these
> regressions would be matrix algebra on chunks of a dataset on the order
> of 1 MB in size.
> 
> I've looked at a lot of configurations on the Dell website ranging from
> a dual-core P4 to a fire-breathing system with a pair of dual-core Xeons
> and all the bells. It would be easy to spend $7-8k and solve our
> problems, but we'd like to be a little more efficient. This is the
> latest version, coming in at about $3000 with a nice monitor. Any comments?
>   Dell Precision Workstation 390
> Intel Core®2 Duo E6600 2.40GHz/1066MHz/4MB L2/Dual-core/VT,
> Reason: Core 2 Duo performance numbers overall are impressive although
> weighted towards multimedia.   This is a mid-range version selected for
> price/performance and the 4MB cache. The 5100-series Xeon is a tempting
> alternative although expensive.
> 
> OS: Genuine Windows® XP Professional, SP2 with Media
> Reason: SAS 9.1.3 is only licensed for Windows and  does not yet run
> under a 64-bit OS.
>   Memory    4GB, 667MHz, DDR2 SDRAM Memory, ECC (4 DIMMS)   Reason:
> Maximum amount of RAM available under Windows XP.
> 
> Hard Drive Configuration C5 All SATA drives, RAID 1, 2 drive total
> configuration  Drives: 80GB SATA, 10K RPM Hard Drive with 16MB DataBurst
> Cache
> 
> RAID 1 is selected for data integrity. I'd do RAID 0 for access speed
> and store the data on a file server, but I've been overridden.
> The 10 k SATA drives are a mid-range alternative between the standard
> 7200 rpm SATA drives and the faster but expensive SAS drives (no
> relation to the software!)
> 
> SAS performance tuning documents suggest that the I/O system is
> generally the bottleneck for SAS programs in practice, although that's
> the one thing that has never limited the particular project that we are
> dealing with here.
> 
> Thanks for taking the time to read this!
> 
> Mike Kenny
> Medical Biostatistics

ATOM RSS1 RSS2