> I don't know what to try anymore...
First rule of bug reporting, which version of the kernel are you using?
Oh, and what type of NFS servers?
Details, details please.
> I'm getting XFS filesystem corruption, I can see this when using ls or
> du. Filenames are corrupted and also their content. I was doing the
> first test on a DELL Precision Workstation with 4 IDE drives but I have
> changed now. So let me explain what exactly I'm doing.
> I have set up a DELL PowerEdge 1400 Server, PIII800 / 256MB /
> ServerWorks CNB20LE. Two U160-SCSI Disks on the first onboard controller
> (AIC-7899). /, /boot and 2x1GB swap are on SoftRAID1 on those two disks.
> Until here, no problem at all, using kernel PR1-PR3. Then I installed
> one Promise Ultra100TX2 IDE controller, connecting 4 IBM 60GB drives. I
> created 1 RAID5 on those 4 drives. I then do a 'sysctl -w
> dev.raid.speed_limit_min=10000' to resync the raid faster, otherwise it
> takes days to sync. Then while syncing, I create an XFS filsystem on it
> and mount it on /home. Now I copy some GB of data from 2 NFS servers
> (while it is still syncing). This is going slow because of high priority
> syncing, but beside that, not problem at all. Later then, after the sync
> has finished and after some reboots, I just made an ls -R /home and
> found out that the filnames were corrupt. I know that what I did is a
> torture for the system, but it should be able to handle such situations.
> Can somebody tell me what could cause the problem. Could it be the
> combination of RAID5 / XFS / syncing / heavy load? Unfortunately there
> is absolutely noting to find in the kernel logs.
> My next steps before giving up:
> - I have installed a second Prosime controller to make sure every IDE
> disk has it's own channel. (Don't blame Promise, I had exactly the same
> prob with the i820 IDE of the DELL Precision 220). Test is running right
> - Configuring the 4 IDE disks as RAID10 and test again. I will loose
> 60GB, but at least we then know that SoftRAID5 with IDE with XFS with
> ... with ... is DANGEROUS(tm).
> - Try with ext2 on the RAID5 :-(
> Thanks in advance for any help