RE: OT: cluster size=0

[chu tes-mail jpl nasa gov (Eugene Chu)]

Speaking of file system problems, I have one that maybe folks on this
list can help solve.

I recently got a API UP2000 with two 833 MHz processors and 3 disk
drives.  A 36 GB drive is used as the boot disk running some recent
dist of RH linux.  Two 73 GB drives were RAID 0'd together to form a
very large volume for data storage.  All drives are on the base SCSI
bus on the mother board, which I think is Ultra-II Fast Wide SCSI.
(This was the way it was delivered, and I haven't had a chance to do
much with it yet.)  If anyone from API is watching, this is the system
you provided to WSM for me to evaluate.

I tried to run a simple file I/O test program to check the performance
of RAID 0 on the same bus, as I've heard of some people getting pretty
good results this way.  The program is very simple; it simply creates a
file, and starts to write a specified buffer size of random data to the
file a specified number of times.  In the standard tests that I run,
the buffer size is 1 MB, mem aligned to 65536 bytes, and I write 1024
of them to create a 1GB file.  My test brought lots of SCSI errors in
the system log file, and the mounted partition became inaccessible.  I
was able to dismount it, but remount attempts gave errors complaining
of unsupported fs type.  fsck got similar errors.

I tried running the same test on the 36 GB root disk, and managed to
complete the run, but only got about 3.3 MB/s transfer rate.  This
seems awefully slow for Ultra-II SCSI.  According to the specs I've
seen for most 36 GB disks, they should be able to run between 20-45
MB/s rates depending on which zone your data happens to have landed,
with average of 30 MB/s.

Right now, the system is down for some reason, and I can't get to it
until Monday.  In the mean time, can anyone provide any pointers on
disk and RAID setup in Linux?  I am most interested in high
performance, but also need reliability (who doesn't?).

thanks in advance,
eyc