NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/sys/apollo/3275

Text File | 1992-08-14 | 3.9 KB | 73 lines

Newsgroups: comp.sys.apollo
Path: sparky!uunet!wupost!gumby!destroyer!terminator!pisa.citi.umich.edu!rees
From: [e-mail] (Jim Rees)
Subject: problem with writing large files from Fortran
Message-ID: <[e-mail]>
Sender: [e-mail] (Usenet Owner)
Reply-To: Antonio Hernandez <[e-mail]>
Organization: University of Michigan IFS Project
Date: Fri, 14 Aug 1992 12:56:47 GMT
Lines: 61

Can anyone help this guy?

Date:         Thu, 13 Aug 92 14:09:14 EST
From: Antonio Hernandez <[e-mail]>
To: Jim Rees <[e-mail]>
Message-Id: <[e-mail]>
Subject:      question

Dear Mr.Rees: I have a question Which I would like to formulate to the
Apollo world. If this is not the appropriate channel please forgive me
and let me know how to proceed. Thanks: Antonio Hernandez.
from: Antonio Hernandez
e-mail: avh at ibma.mpa.ipp-garching.mpg.de
subject: question to the apollo world

 Question: I have an Apollo DN10000 at the Simon Bolivar University
(USB) in Caracas-Venezuela. It runs under SR.10.2.p and has one
multidisk (w0:0+w1:0) comprising two 700Mb units in a sector-striped
configuration plus a second multidisk (w0:1+w1:1) of two 370Mb units
with the same sector-striped configuration. It also has two processors
and 64Mbytes of RAM:  Let's call this machine "USB" (Fortran Compiler
ftn.v.10.8p).
I also work with a DN10000 at the Max Planck Institute for Astro-
physics in Munich Germany in a colaboration project. This station,
which we can call "MPA", is a one processor machine with a multidisk
(w0:0+w0:1) of two 700Mb units in a cylinder-striped configuration,
with 32Mb RAM and running under SR.10.1.p. (Fortran Comp.:ftn.v.10.7p)
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++=======>
The problem is as follows: I run MUNICH code (fortran code for
molecular calculations) on the MPA station without any trouble. This
particular code opens a large file for reading (about 100MB)and
simultaneously opens a file for writing the processed data: this
second file will be also very big at the end of the program  (some
200MB). The MPA mchine does mostly I/O and expends little CPU time on
this molecular code.
--When I run the same code on the USB machine, a strange thing happens
when half of the output file is written: You can hear large disk
activity but the size of the output file doen not change, i.e, the
machine is writing data to some temporal file (swapping area?) and
from there to god knows where!, but it never writes the final data
to the output file. It seems like there is a limit on the number of
blocks this machine can move, or it lost the address of the last
record read on the first file. Both read and write files are sequential
unformatted with a record length of 1400 integer words.---
I have tried recompiling but when the size of the problem is small the
code runs without problems (small input output files of about 40Mb) and
when the size increases (input output files of about 80Mb) it breaks.
Once I could make it run by running the code on the first multidisk
(I/O also in the first multidisk) and I move all paging (swapping) to
the second multidisk: On other words, I keep the operating system and
the code running on the first multidisk and turn off paging there by
changing the name of the swapping device on /dev directory, and
define new OS paging area only on the on the second multidisk with the
/etc/swap command.
All this is very strange because here in Germany I do not seem to have
any problem with the size of the problem: Right now I am running a
very large case that handles I/O files of 300Mb each!. Is there a
problem with the RS.10.2.p operating system? or with the way it
handles swapping, because the /etc/swap command do not even exist under
RS.10.1.p operating system at the MPA machine. Is there a hardware probl
                       ****PLEASE HELP****Antonio Hernandez.