Date:         Sun, 5 Oct 1997 20:47:06 -0700
Reply-To:     "Allen J. Michielsen" <amichiel@EARTHLING.NET>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         "Allen J. Michielsen" <amichiel@EARTHLING.NET>
Organization: No Organization Identified Anywhere Near Here
Subject:      Re: Comments on SAS Performance in NT and UNIX?
Content-Type: text/plain; charset=us-ascii

Jianmin Liu wrote:
>
> Hi All,
>
> I'd like to have some info. or comments on SAS performance
> in NT and UNIX.  I understand SAS procedures are subject to
> 2.1 GIG limitation of dataset size in NT environment (imposed
> by MS).  When I process a data set with 1.9 gig in NT, SAS becomes
> unstable and get some unreasonable errors.  I had to split the
> data set in two subsets.  Then everything is fine.
>
> SI technical staff agreed that is might be due to the 2.1 gig
> limitation imposed by Microsoft (Note, SAS only
> supports a few procedures in NT when the data set is
> greater than 2.1 gig.  But my experience suggests that
> even when the size is getting close to 2.0, you might
> be in trouble.  Again, according to SI staff, this
> limitation is imposed by MS.
>
> When I used dde in NT to read MS Access data into SAS, a
> similar limitation imposed by Microsoft was encountered.  As
> a result, we licensed ODBC driver from SAS for MS Access
> interface (of course extra fee).
>
> I guess in UNIX there is no such a limitation.
> Personally I perfer UNIX platform running SAS, not only
> because of the limitation mentioned above.  Running
> time is also a big concern.
>
> In the company I'm working at, we are now seriously considering
> having a UNIX server.  NT is far from ready  for dealing with
> large data set, in my opinion.
>
> Any comments would be appreciated.
>
> Thanks.
> JL
 
The 2 GB limit of a disk under NT is only real for pysical drives. I had
a system with a 12 GB NT volume created by disk striping.  Multiple
drives, each storing only part of the data.
Not many Unix versions do not have a similiar physical volume storage
limit.  Most do not support striping (yet).
In order to process a 1.9GB data set, you will need 1.9GB+ for the temp
working space. My experience with large datasets on NT is that as
physical resources are expired, processing time increases tremendously.
Unlike a mainframe environment, where more resources are available but
held in reserve for other jobs, a NT/Windows/Unix environment holds
nothing in reserve.  This allows the system to much more easily hard
crash by errant user jobs.  Large datasets are best 'processed' on NT by
using procedures only.
However, funny errors and problems are a hallmark of Windows operating
systems in general when running out of system resources.  Unix generally
is more stable when running out of system resouces, and generally at
least a error message is written to the system log so you know what
happened before you died.