I wrote a simple C-Program to get an overview about speed of a disk
subsystem.
If you open (see "man 2 open") a file with the O_SYNC option:
"O_SYNC The file is opened for synchronous I/O. Any writes
on the resulting file descriptor will block the
calling process until the data has been physically
written to the underlying hardware. See RESTRIC-
TIONS below, though."
...you are not depending on cache anymore. use the device raw (dont mount
the disk nor use mkfs). you should be able to get the real speed of the
hardware.
But nobody use pure hardware. It's always a question about OS, driver,
application and hardware.
hm...and that's the point where "disk speed" programs run out of reality.
cu,
Martin Lonkwitz
-----Urspr�ngliche Nachricht-----
Von: Ferguson, Neale [mailto:[EMAIL PROTECTED]
Gesendet: Freitag, 12. Dezember 2003 22:18
An: [EMAIL PROTECTED]
Betreff: Re: DASD Performance problem
I like iozone. You can cause it to do direct I/O (i.e. bypass Linux's
cache).
-----Original Message-----
Although everyone should have his own preferences for benchmarks, I
don't think dbench is a very good one to do I/O benchmarking. It is more
an efficiency test for the filesystem. One of interesting things about
dbench as an I/O benchmark is that it does not do *any* I/O to disk in
some configurations. There's so many variables in that equation that it
is very hard to compare results, especially between different platforms.
Rob
