Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!sol.ctr.columbia.edu!cica!news.cs.indiana.edu!ux1.cso.uiuc.edu!ux1.cso.uiuc.edu!uxa.cso.uiuc.edu!rjtg0209
From: rjtg0209@uxa.cso.uiuc.edu
Newsgroups: comp.lang.fortran
Subject: Slow unformatted I/O
Message-ID: <50700005@uxa.cso.uiuc.edu>
Date: 4 Dec 90 15:06:00 GMT
Lines: 37
Nf-ID: #N:uxa.cso.uiuc.edu:50700005:000:1419
Nf-From: uxa.cso.uiuc.edu!rjtg0209    Dec  4 09:06:00 1990



My programs do a lot of unformatted I/O. I hoped that such an I/O in
FORTRAN is done close to the disk speed but this is not necessarily
true. On some workstations, the I/O speeds have been a magnitude
slower.  I wrote the following program to benchmark the unformatted
I/O:

       integer x(1000)
       open(1,file='scr',form='unformatted')
       do i=1,400
          write(1)x
       enddo
       rewind 1
       do i=1,400
          read(1)x
       enddo
       end

The problem is that the wall clock time for this program is much longer
than the CPU time on several computers that I tested. Foe example, the
CPU on SPARSTATION 1 shows the transfer rate of 2.5 Mbytes/s, while the
wall clock time on the unloaded machine shows only 300 kbytes/s.
Perhaps this is associated with the buffer size allocated for each open
file. The real speed of loaded Cray-2 which has a 256 kbyte buffer is
close to 15 Mbytes/s. The speed of my PC 386/33MhHZ with 640 Mbytes
ESDI disk, using the program above and the Microsoft 5.0 compiler is
200 kbytes/s. However, if one reserves a 32k buffer by the blocksize
option in the open statement, the speed increses four-fold, to 800
kbytes/s.  Does anybody have any experiences with these kinds of
problems? Which compilers have options to increase the buffer size or
do it automatically?

Please send your response to ignacy@uiucvmd.bitnet 

Ignacy Misztal 
University of Illinois


Brought to you by Super Global Mega Corp .com