Date:         Tue, 17 Apr 2012 14:44:36 -0400
Reply-To:     "Gilsen, Bruce F." <bruce.gilsen@FRB.GOV>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Gilsen, Bruce F." <bruce.gilsen@FRB.GOV>
Subject:      Surprising platform results: any thoughts?

We ran the following code the other day with Version 9.3 on Linux and mainframe:

%macro all;
  %let inc=200;
    DATA ONE;
      %DO I=1 %TO &inc;
        %DO J=1 %TO 1000;
          X&I=&J;
          Z="AAA";
        %END;
      %END;
    RUN;
%mend all;
%all;

The results were pretty amazing to me.  CPU time / real time in seconds:
Linux: 5.42 / 5.42
Mainframe: 105:44 / 202.71

I ran with 9.1.3 and the results were the same.  Focusing on the CPU time,
the mainframe was ~20 times slower, which just seems hard to believe.

The other SAS consultant took a quick look at Mike Raithel's seminal book on
mainframe tuning and nothing seemed out of whack.  SAS Institute looked at
similar code last week and it ran about equally fast on mainframe and Linux,
but, they have a faster mainframe.  We are still talking to them.

Machine details:
Mainframe: IBM Z9BC Model O02, 2 126.5 engines for 253 MIPS
Linux: 40 processors (4 10-core processors), 256GIG memory server, E7-4870
processors, 1066 megahertz memory, 7 terrabytes RAID 6 storage, Kernel
number 2.6.18, Red Hat 5

Has anyone gotten similar results at their site?  Any ideas if this really
makes sense?

===

Bruce Gilsen
speaking only for myself (and my fantasy baseball team, which is
trying for its 5th title in 24 years in 2012, though if Tim Lincecum doesn't
shape up I'm screwed)