Date: Thu, 11 Mar 1999 14:09:28 -0500
Reply-To: "Steven E. Stevens" <sstevens@COMCAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From: "Steven E. Stevens" <sstevens@COMCAT.COM>
Subject: Re: SAS vs. Oracle
Content-Type: text/plain; charset="us-ascii"
Jesse, Mohanty:
Can't dispute your point # 2. below - however, I have a bone to pick about
# 1. (& not just with you).
I've heard Oracle can perform joins much faster than SAS, etc., etc., time
& time again on this list, as general "common knowledge," & even from SI
personnel (who shall remain anonymous, of course). Well, some former
colleagues (who are probably reading this now) performed a carefully
controlled performance test of SAS vs. Oracle in decision support-style
queries & guess who won? Before I reveal that, let me assure you that we
used identical hardware & OS (HP K400-class running HP-UX 10.x), identical
database design & content (Oracle V7.3X database tables & indexes vs. SAS
6.12 datasets & indexes, total data size ~= 8 GB), and identical queries
(Oracle SQL vs. SAS proc sql). In fact, one of the queries had even been
optimized for Oracle using EXPLAIN PLAN & tkprof.
Bottom line, SAS creamed Oracle & I mean, REALLY creamed it (avg speedup
SAS vs. Oracle ~= 55%)! No one was more surprised than me since I also
believed the line propounded in # 1. below. Now, I don't believe SAS will
ALWAYS outperform Oracle on read-only queries (too many factors at play
here - hardware & OS configuration, database design, data placement,
etc...), but its difficult to argue with empirical results for a specific
database. My suggestion (if you have the time & resources) is to conduct
experiments on representative data to see for yourself rather than accept
"folk wisdom" (apologies in advance Mohanty, I'm not trying to pick on you
here).
Regards,
S.Stevens
>X-Mailer: Internet Mail Service (5.5.2232.9)
>Newsgroups: bit.listserv.sas-l
>Date: Wed, 10 Mar 1999 12:11:01 +0530
>Reply-To: "Mohanty, Debasmit (CAP)" <Debasmit.Mohanty@GEIND.GE.COM>
>Sender: "SAS(r) Discussion" <SAS-L@VM.MARIST.EDU>
>From: "Mohanty, Debasmit (CAP)" <Debasmit.Mohanty@GEIND.GE.COM>
>Subject: Re: SAS vs. Oracle
>Comments: cc: Jesse Chittams <jchittam@CCEB.UPENN.EDU>
>To: SAS-L@VM.MARIST.EDU
>
>Jesse:
>
>You have mainly talked about 2 issues:
>
>1. Merging large datasets
>- Here you should try to leverage on the strength of Oracle. Since the
>Oracle database would have indexes / partitioned indexes for all the joining
>keys, merging would be faster in Oracle compared to that in SAS.
>
>2. Deriving variables on complex criteria
>- For this, I would say SAS is better as it has lot more flexibility
>than in Oracle.
>
>To summarize, I would say one should do the DATA PREPARATION activities (
>querying / joining / filtering/ summarization ) in Oracle
>Where as use SAS for all DATA TRANSFORMATION activities (derived variables
>etc )
>
> Any other thoughts ?
>
>-Debasmit
>
>g GE Capital
>_________________________________________________
>Debasmit SR Mohanty
>Assistant Manager - Systems
>GE Capital Card Services - I M
>A-1, Golden Enclave, Airport Road,
>Bangalore - 560 017, India
>
>Phone : +91-80-527 5535 Ext 107
>Mobile : +91-98440-25340
>Fax : +91-80-527 5930
>Internet : Debasmit.Mohanty@geind.ge.com
>URL : http://www.angelfire.com/me/debasmit/
><http://www.angelfire.com/me/debasmit/>
>Disclaimer: The opinion herein expressed are of mine and
>Doesn't reflect that of my employer
>_________________________________________________
>
> ----------
> From: Jesse Chittams [SMTP:jchittam@CCEB.UPENN.EDU]
> Sent: Wednesday, March 10, 1999 2:05 AM
> To: SAS-L@UGA.CC.UGA.EDU
> Subject: SAS vs. Oracle
>
> Hello,
>
> Has anyone compared SAS and Oracle on UNIX regarding, merging large
>data
> sets (millions of records and hundreds of variables), and deriving
> variables base on complex criteria?
>
> Thanks,
> -Jesse Chittams
>
>
| 
