m and n on 750 w/ RA81's
Ed Bryant
ed at sfucmpt
Sat Nov 26 03:41:34 AEST 1983
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
More information about the Comp.unix.wizards
mailing list