On Thu, 23 Nov 2006 08:02:39 -0800, Sekhar <ckalisetty@GMAIL.COM> wrote:
>OK. I fixed the memory problem by using -memsize max option. Now I have
>different problem. I wanted to keep observations only when they are in
>both datasets. That's why I used "if rc=0 then output". But by doing so
>I got onlly 30 obs in the output dataset. How can I implement the
>condition that output dataset only will have observation with non
>missing open date? thanks in advance.
Please provide aready-to-use samples of WORK2 and COMBINE, as well as a
mockup of the expected results.
>On Nov 17, 4:01 pm, EvilPettingZo...@AOL.COM (Ken Borowiak) wrote:
>> On Fri, 17 Nov 2006 12:43:07 -0800,Sekhar<ckalise...@GMAIL.COM> wrote:
>> >Looks like I have memory problem.
>> >Here is my code.
>> >***** first declare a hash table for the small dataset work2 ******;
>> >data test;
>> >declare hash h_small ();
>> >***** defien this hash table ******;
>> >length prod_acc open_dt losbal 8.;
>> >rc= h_small.DefineKey ("prod_acc");
>> >rc= h_small.DefineData ("open_dt","losbal");
>> >rc= h_small.DefineDone ();
>> >**** fill this hash table *******;
>> >do until(eof_small);
>> >set work.work2 end=eof_small;
>> >rc = h_small.add ();
>> >**** access the hash table *****;
>> >do until(eof_big);
>> >set work.combine end=eof_big;
>> >open_dt=.; losbal=.;
>> >rc=h_small.find ();
>> >When I run this with 10,000,000 obs I get the following error message.
>> >ERROR: Hash object added 8126448 items when memory failure occurred.
>> >FATAL: Insufficient memory to execute data step program. Aborted during
>> >the EXECUTION phase.
>> >But when I run the following code to check memory no error.
>> >%let n = 35000000;
>> >data _null_;
>> >array k[&n] _temporary_;
>> >Should I look for bigger memory. ?You can use your memory resources a
bit more selectively if you hash a
>> record pointer (rather than 2 satellite fields) and using the POINT= on the
>> SET statement to retrieve the satellite records.
>> See the latter portion of the Paul Dorfman and Lessia Shajenko's SUGI 31
>> paper "Crafting Your Index: ".