Disk I/O

[Christopher Reid Palmer <reid@pconline.com>]

Hi there,

I'm curious about how well my storage subsystem can perform, so I grabbed
the benchmark program Bonnie and fired it up. I'd like to compare results
with the rest of y'alls.

My hardware:
============
K6 233, 66MHz bus (I think)
Mylex/BusLogic Flashpoint Ultra SCSI (narrow) host adaptor
IBM ?? model 7200 RPM SCSI u2w, /dev/sda
    (with wide->narrow adaptor so it'll work with my SCSI card)
Seagate Hawk 5400 RPM SCSI II, /dev/sdb
64MB RAM
dusk:~ $ df
Filesystem           1k-blocks      Used Available Use% Mounted on
/dev/sda2              2966789    687679   2125690  24% /
/dev/sda3              5560730   1964229   3308536  37% /usr/local
/dev/sdb1              1018298    315351    650336  33% /backup



My results:
===========
dusk:~/src/Bonnie $ ./Bonnie -s 100	# 100MB test file
...
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   849 86.9 12357 60.0  4009 32.1  1863 81.6 10625 26.5 144.8 4.2

Let's run it again as root just for fun:
========================================
dusk (~) # ./Bonnie -s 100
...
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   978 98.1 14486 64.8  5198 40.5  2116 91.3 12535 32.8 170.6 4.6

Block seems to be pretty good, everything else sucks. I'm sure I could be
getting more out of that disk if I had a wide SCSI adaptor; sticking it on
a narrow bus is sadness. I'd have to reboot to check and be sure, but I
think it's set for asynchronous transfers on that SCSI ID.

I ran this test while doing normal things like reading the E10k article in
Netscape (thanks for the link Carl!) and checking my email. It's useless
to measure the performance of a machine sitting still, because machines
don't sit still.

What worries me is the huge CPU eating rates. The Commentary doc that
comes with Bonnie suggest that's a fault of Unix filesystems. I'm told by
a knowledgeable party that ext2 is no better than traditional Unix
filesystems in this regard.

Here are my results using my backup disk, a 5400 RPM Seagate SCSI II:
=====================================================================
dusk (/backup) # ./Bonnie -s 100
...
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   966 97.3  3973 13.0  1575 10.6  2053 88.8  1871  4.5  85.4 2.5

Char seems to be roughly similar--couldn't really get much worse, could
it--but block is much lower, as you'd expect.

Both at once!:
==============
dusk (~) # ./Bonnie -s 100
...
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   495 49.7 10714 44.9  3144 24.4  1568 75.1 10184 25.4 103.7 3.7

--
dusk (/backup) # ./Bonnie -s 100
              -------Sequential Output-------- ---Sequential Input-- --Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
          100   494 49.7  3817 13.3  1426 11.2  1764 79.3  4149 10.3  99.5 2.6


Both disks performed in parallel not too much less well than when the 
benchmark is run on each disk in series. That's the advantage of SCSI.
Also, there was some time in the end where only /dev/sdb1 (/backup) was
being taxed; the faster disk finished ahead of the slower one.

Note the weirdness in sequential block inputs for /dev/sdb when run in
series compared to parallel--it's over *twice as fast* when run in
parallel! This must be some kind of anomaly. I'd have to run the whole
series of tests many times and take an average. Perhaps a shell script is
in my future.

Any comments? Anyone wanna post their results?


--
Christopher Reid Palmer : www.innerfireworks.com