Article View: comp.unix.cray
Article #253atexpert predictions and actual performance on the Cray-2
From: Hemant Dandekar
Date: Fri, 19 Apr 1991 15:34
Date: Fri, 19 Apr 1991 15:34
52 lines
2965 bytes
2965 bytes
Hi,
I am trying to run a program written in Fortran on a Cray-2 running
UNICOS 6.1 and CF77.4.0.3 under autotasking. Using the option -Wu"-p"
and running my code in time share mode I generate the output which can
be used by 'atexpert' a autotasking tool. The 'atexpert' tool predicts
the speedup of the code in 'standalone' mode based on the information it
is fed. However, I get very poor speedup compared to the predictions
when I run my code on a 4 processor Cray-2 at NCSA. Here is a comparison
between predicted and actual values.
# of CPUS Speedup (atexpert) Speedup(Actual)
--------------------------------------------------------------------
1 1.0 1.00
2 1.9 1.27
3 2.8 1.32
4 3.6 1.33
--------------------------------------------------------------------
Before I run my code I set my MICRO_NICE value to 0 and my NCPUS value
(these are shell variables) to 1, 2, 3 or 4. I do not set my
MICRO_TIMEOUT variable to a large value as suggested by Dick Steimann,
of CCN Cray Systems group. However, I don't think any of my slave CPU's
remain idle as the breakup of CPU time for a typical run is as follows:
NCPUS = 3
1 CPU 83.31s
2 CPU 45.76s
3 CPU 26.40s
out of a total CPU time of 155.47seconds.
An obvious way to check things out will be to make another dedicated run
with setting the MICRO_TIMEOUT variable to a large values such as 10000.
However, I cannot make another run for another week. Meanwhile, I would
be curious to know about the experiences of other users on the net with
regards to autotasking on the CRAY-2 under UNICOS 6.0+ and their
experience with 'atexpert' predictions and actual performance.
Thanks,
hemant
*********
BITNET: V092QGHG@UBVMS.BITNET |Hemant W. Dandekar
INTERNET: hemant@eng.buffalo.edu|303, Furnas Hall, SUNY at Buffalo
u10978@u2.ncsa.uiuc.edu|Amherst NY 14260 (716)636-2631(work)
--
----------------------------------------------------------------------------
Bitnet: v092qghg@ubvms.bitnet | Hemant W. Dandekar
Internet: hemant@asterix.eng.buffalo.edu| 303, Furnas Hall, Chem. Eng.
(716)636-2631 | SUNY at Buffalo, Buffalo NY 14260
Technical Details
Message-ID:
<71860@eerie.acsu.Buffalo.EDU>Path:
arc1-ca.pugleaf.net!utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!ub!acsu.buffalo.eduShow Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!ub!acsu.buffalo.edu From: hemant@acsu.buffalo.edu (Hemant Dandekar) Newsgroups: comp.unix.cray Subject: atexpert predictions and actual performance on the Cray-2 Summary: significantly lower speedup than predicted on a Cray-2 Keywords: autotasking, atexpert, speedup, MICRO_TIMEOUT Message-ID: <71860@eerie.acsu.Buffalo.EDU> Date: 19 Apr 91 15:34:43 GMT Sender: news@acsu.Buffalo.EDU Organization: SUNY Buffalo Lines: 52 Nntp-Posting-Host: asterix.eng.buffalo.edu