Date: Wed, 28 Jan 2009 09:09:01 -0800
Reply-To: "gss.sas" <gss.sas@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "gss.sas" <gss.sas@GMAIL.COM>
Organization: http://groups.google.com
Subject: Re: Performance issues with 9.2
Content-Type: text/plain; charset=ISO-8859-1
On Jan 28, 1:06 am, snoopy...@GMAIL.COM (Joe Matise) wrote:
> Proc Sort is appropriately faster, so it's not the CPU at least in that
> instance. I'd wonder if the macro was not written properly for the new
> server - errant libnames, or something. Hard to tell though without knowing
> the code, I suppose.
>
> Does a large datastep similarly run faster, or is it slower? If proc sort
> is faster, and datastep is faster, then I have to think SAS is doing fine at
> least generally. Could be a particular product (test each of the types of
> proc you use in this large macro) or could be something in your SAS code
> that is slowing it down.
>
> -Joe
>
>
>
> On Wed, Jan 28, 2009 at 1:11 AM, gss100 <gss...@gmail.com> wrote:
> > On Jan 27, 3:57 pm, art...@NETSCAPE.NET (Arthur Tabachneck) wrote:
> > > Hardware issues are out of my league, but I do have a question. Do the
> > > data files all reside on the new server?
>
> > > When I first went to a multi-processor SAS server, the first thing I had
> > > to do (for similarperformanceissues) was move all of the data onto the
> > > server.
>
> > > I don't have either Vista or9.2, thus don't know if they have added
> > > anything new to the equation.
>
> > > Art
> > > -------
>
> > > On Tue, 27 Jan 2009 15:44:55 -0800, gss.sas <gss....@GMAIL.COM> wrote:
> > > >Thank you guys for your input.
>
> > > >Here is an update:
>
> > > >1. The CPU is not getting throttled, and the DDR memory is in the
> > > >right slots (all 4 slots have 2 GB each).
> > > >2. When we run Winzip or Winrar, it is at least 200% faster on the
> > > >Vista q8200 machine with 8 gb DDR2 compared to Athlon 4200+ with 4 GB
> > > >DDR.
> > > >3. If we do a proc sort of a large dataset, the Vista machine is at
> > > >least 3 times faster.
> > > >4. Here is the curious part: when we run a large macro, the Vista
> > > >machine starts off faster, but then slows down so that after a few
> > > >hours, it is considerably slower. The system CPU time in the log is
> > > >still faster for the Vista machine but the real time and CPU user time
> > > >are both slower.
> > > >5. If we start a new macro, the same thing happens - it starts
> > > >faster, and then slows down.
> > > >6. The difference is considerable : last night, a large macro
> > > >program took 5 hours on the old XP machine, but 9 hours on the new
> > > >Vista.
>
> > > >There is something seriously wrong here. I think it is software
> > > >related.
>
> > > >Suggestions?- Hide quoted text -
>
> > > - Show quoted text -
>
> > We migrated the new files to the x649.2system with proc migrate. Is
> > this what you mean?- Hide quoted text -
>
> - Show quoted text -
Joe, it is the same identical macro. No problemns with libnames etc -
no errors. Everything runs. Same identical code on both machines,
same datasets.
What is curious is that the Vista machine starts off considerably
faster, both in system CPU time, user CPU time and real time, and then
after a while slows down. CPU time is still fast, but real time slows
down. The performance hit is considerable - 5 hours vs. 9 hours, so
this is quite a serious issue.
When we start a new macro, the same thing happens - it starts off
fast, and then slows down.
It is not Vista / CPU / memory /hard drive related, because the Vista
computer runs much faster in all other tasks including Winrar/Winzip
operations or on Sandra SisSoft benchmark. There is something strange
going on with SAS 9.2 in a x64 environment.
Other suggestions?
| 
