We have a dual PII with 8 4.3GB Cheetahs in two uw scsi channels. I
used to run it with kernel 2.2.3, a root disk, spare disk and a raid5
partition composed of 6 disks. The raid5 partition worked otherwise
well, but within 6 months it lost its superblock three times, once in
a UPS failure, twice from no apparent reason. I had to do mkraid -f -o
on it. I grew tired of this and decided to try a newer kernel and raid
version and 0+1 raid.
So I installed kernel 2.2.11, the 19990824 kernel patches and
corresponding raid tools (0.90). I configured two raid0 partitions,
each with three disks, and then one raid1 partition on top of
these. It seemed to run nicely, benchmarks were good and no
glitches. But in a few days the system began showing signs of running
out of memory. It has 512MB RAM, but even a small load would make it
swap. Ps output showed no memory hogs, so it seems it was some part of
kernel that was leaking. The kernel configuration didn't have anything
special except raid. I had all raid options enabled in the
kernel. After rebooting everything was fine, but it deteriorated in a
week or so to unusable.
I tried to upgrade to kernel 2.3.39, older raid and two raid0
partitions without raid1, but could find no raidtools version that
would work with the 2.3.39 kernel. Am i supposed to patch the latest
2.3 kernels? With what, and what raidtools to use with them? The raid
and md HOWTOs are very old and even the raidtools docs are vague.
How about kernel 2.2.14? It wouldn't patch with the 19990824 stuff.
I find the interplay of kernel, kernel patch and raidtools versions
very confusing. What is considered a stable combination, preferably
with 2.2.14 or latest 2.3 kernels?
Janne Himanka
