Date: Thu, 11 Oct 2001 19:15:27 +0000
Reply-To: Puddin' Man <pudding_man@POSTMARK.NET>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Puddin' Man <pudding_man@POSTMARK.NET>
Subject: Re: Merge w/o sort first
Content-Type: text/plain; charset="iso-8859-1"
IanD wrote:
> With reference to message 5 in thread, my understanding of SQL was
> that it used less memory to sort a dataset.
In doing some testing, I find that SQL uses somewhat more
memory than PROC SORT.
> I believe that PROC SORT
> needs approx 3x the size of the dataset availabe to be able to sor
> it.
The '3x' refers to sort-work space on disk. I believe that any
standard method of sorting implemented with SAS requires
something approximating the '3x'. This excludes TAGSORT
and TAGSORT-related approaches, etc.
> PROC SQL requires less.
I have no evidence of this. Can you test?
> I remember reading this a long time ago
> but, obviously, have absolutley no evidence to back this up.
If anyone
> could enlighten me I would be obliged. Thanks for being genle
with a
> newbie.
I would recommend running your own tests. I am finding that SQL
generally uses more system resources than PROC SORT. Results
may vary across OS's.
SQL sort performance on the mainframe appears to be particularly
inferior, possibly owing to it's inability to access system sort
utilities (i.e. DFSORT, SYNCSORT) that are available to PROC SORT.
My OS/390/DFSORT results are consistent with Paul's. A PROC SQL
sort appears to be using many times the EXCP's used by PROC SORT
--> DFSORT. "Something Is Bad Wrong Here."
In fact, sort performance with SQL on my OS/390 is so miserable that
I intend to "go the extra mile" to avoid using it in the future.
Skoal,
Puddin'
*******************************************************
*** Puddin' Man *** Pudding_Man@postmark.net *****
*******************************************************;
MadMan expounds on events of recent years:
"Lordy me, but we sho' could use a big ol' passel of domestic
policies whot don't get us Bloody Murdered!"
Pauses a few seconds, bobs his eyebrows twice, then:
"Lordy me, but we sho' could use a big ol' passel of foreign
policies whot don't get us Bloody Murdered!"
Iffen he sez insane stuff like dat, he's _gotta_ be
100% Stone Crazy. Right? Right??
> Cheers
>
> IanD
>
> Paul.Dorfman@BCBSFL.COM (Dorfman, Paul) wrote in message news:
<17DA667110692-01@mail.bcbsfl.com>...
> > Puddin',
> >
> > My results were the same as yours on PC: SQL was slightly
slower (both clock
> > and cpu). On the mainframe, SQL was slower at the ration
of 3:1 (both
> > again).
> >
> >
> > > Paul,
> > >
> > > Outperformed SQL by what criteria? cpu? excp's?? wall clock???
> >