Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!linus!decvax!microsoft!uw-beaver!ubc-visi!sfucmpt!ed
From: ed@sfucmpt (Ed Bryant)
Newsgroups: net.unix-wizards
Subject: m and n on 750 w/ RA81's
Message-ID: <169@sfucmpt.UUCP>
Date: Fri, 25-Nov-83 12:41:34 EST
Article-I.D.: sfucmpt.169
Posted: Fri Nov 25 12:41:34 1983
Date-Received: Thu, 1-Dec-83 04:26:39 EST
Lines: 73

And now, more in the m and n saga:

We have a VAX 750, 4Mb memory, and a couple of ra81's on a uda50 controller,
running Unix 4.1b.  Since it wasn't doing anything for a couple of days,
I ran the following script:

#!/bin/csh
echo "test of disk m and n factors"
date
foreach x (1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17)
    foreach y (100 256 300 327 357 364 512 714 728 1024)
        echo "m n: $x $y"
        fsck -s$y\:$x /dev/ra1g >/dev/null
        mount /dev/ra1g /mnt
        echo -n "    Writing:    "
        time sh -c "dd if=/usr/bigfile of=/mnt/test bs=10240 >/dev/null 2>&1"
        echo -n "    Reading:    "
        time sh -c "dd if=/mnt/test of=/dev/null bs=10240 >/dev/null 2>&1"
        echo -n "    Reading:    "
        time sh -c "dd if=/mnt/test of=/dev/null bs=10240 >/dev/null 2>&1"
        rm -f /mnt/test
        umount /dev/ra1g
     end
end
echo "test complete"
date

According to the ra81 manual the n should be 364 calculated as follows:
7 surfaces @ 2 heads per surface @ 52 512 byte sectors per track =
728 sectors per cylinder, or 364 file blocks per cylinder.

The results I obtained are:
---------------------------------------------------------------------------
  Results of disk tests on a Vax-11/750, ra81, with uda50 controller
  m = disk sector skip factor, n = file blocks per cylinder
  Average read time (in seconds, 2 tries) of a 10 Mbyte file
 m\n |  100 |  256 |  300 |  327 |  357 |  364 |  512 |  714 |  728 | 1024 |
----------------------------------------------------------------------------
  1  |  168    168    168    167    167    167    168    168    168    168 |
  2  |  169    170    169    169    170    170    170    170    170    170 |
  3  |  171    171    172    171    171    171    171    171    171    171 |
  4  |  116    115    115    114    115    115    115    114    115    115 |
  5  |  113    112    111    112    112    111    111    111    111    111 |
  6  |  118    117    117    117    118    116    115    116    117    115 |
  7  |  110    108    106    107    109    107    107    106    107    107 |
  8  |   89     87     88     88     89     88     88     87     89     87 |
  9  |   84     83     84     83     83     84     83     83     83     84 |
 10  |   92     90     88     87     86     87     87     87     87     87 |
 11  |   96     95     96     94     94     96     96     96     96     96 |
 12  |  104    102    102    103    101    104    104    104    103    104 |
 13  |  111    110    110    113    109    111    111    111    111    111 |
 14  |  118    117    117    120    117    119    119    119    120    119 |
 15  |  126    126    125    128    124    127    127    126    127    127 |
 16  |  134    134    133    135    132    134    135    134    134    135 |
 17  |  141    141    140    143    141    142    142    142    142    142 |
----------------------------------------------------------------------------
the read time is the elapsed time, since the system time and user time
are the same in each case. The theory is, I believe, that system time does
not include waiting for the disk.
These results are puzzling, since it appears that the choice of n is
irrelevant, but the choice of m is. This appears to be a contradiction
to the Unix manual! Perhaps the uda50 is smarter than average?
These results are fairly accurate, I think, since there are no terminals
connected to the VAX, and no other system activity.

If and when 4.2 arrives, I propose to run these tests again,
and if the results are any different, I will post them.

Ed Bryant
Laboratory for Computer and Communications Research
Simon Fraser University
(604-291-4430)
... ubc-vision!sfucmpt!ed