Date: Tue, 24 Jul 2007 08:42:32 -0700
Reply-To: Jack Hamilton <jfh@STANFORDALUMNI.ORG>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Jack Hamilton <jfh@STANFORDALUMNI.ORG>
Subject: Re: SAS Product Selection
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
What I am trying to do: load a large Hash Object (2+ million rows).
I hope some future version of SAS will allow hash objects that aren't
stored entirely in memory. I think FREQ and SUMMARY once had the same
restriction, and were later revised to use disk spill files.
Robert Bardos wrote:
> Really! However I was careful enough not to claim that I specify that much
> on a JOB card. I wrote '1 to 2,5 GB for sorting' and that's what I do. I
> specify some 256M or 512M on the jobcard and add a DFSPARM DD statement
> which specifies things like HIPRMAX=2000,MOSIZE=0.
> This is one of the messages in the log (DFSORT's SYSOUT):
> ICE180I 0 HIPERSPACE STORAGE USED = 2'047'656K BYTES
> (formatted for readability). When I watch the region (memory) go up and
> down during these long running jobs I do see values on SDSF/EJES that
> correspond to these 2GB plus the amount specified (and tolerated) on the
> JOB or EXEC statement.
> Had posted my findings in this regard on March 5, 2007
> Just for the sake of experiment I ran an IEFBR14 with REGION=2048M on the
> jobcard an got exactly the same message as you.
>> -----Ursprüngliche Nachricht-----
>> Von: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU]Im Auftrag von
>> Jack Hamilton
>> Gesendet: Dienstag, 24. Juli 2007 16:41
>> An: SAS-L@LISTSERV.UGA.EDU
>> Betreff: Re: SAS Product Selection
>> Really? I can't even ask for 2GB of memory - "// JOB REGION=2048M"
>> gives me a JCL error saying that 2048M is more than I can even possibly
>> get: "IEF638I SPECIFIED NUMERIC EXCEEDS MAXIMUM ALLOWED ON THE JOB
>> Robert Bardos wrote:
>>>> Jack Hamilton
>>>> On the other hand, the mainframe is not the place to run
>>>> memory-intensive programs. I can get 3 GB of (mostly
>> virtual) memory on
>>>> my desktop, which is now several generations old. I can't
>> get even half
>>>> of that on our mainframe.
>>> Well, the 'your' mainframe certainly applies. I'm running MICS SAS jobs
>>> every night where up to some six jobs run in parallel each
>> requiring (and
>>> getting) some 1 to 2,5 GB of memory for sorting (limit set by
>> me; I never
>>> felt greedy enough to test the actual limit). And this is only
>> part of the
>>> nightly batch load the system handles quite happily.
>>> Memory on modern mainframes running all kinds of funky applications is
>>> measured in significantly higher numbers than it used to be a few years
>>> Robert Bardos
>>> Ansys AG, Zurich, Switzerland