At 10:25 18.05.99 +0200, Ingo Molnar wrote:
>Currently stock 2.2.9 doesnt really work well with RAID1/RAID5 (reason yet
>unknown, Doug Ledford is investigating the issue). This could either be
>the buffer-cache changes in 2.2.8, or a genuine RAID problem. 2.2.7 (with
>the small patch i've sent to this list that resolves a small reject) works
>fine with the latest release.
Did some more experiments:
The problem first occured with the changes to the scheduler introduced in
2.2.7-ac3 (2.2.7-ac2 runs fine).
These changes were integrated into 2.2.8 and live on in 2.2.9.
The wait_on_buffer calls used in md_do_sync lock after a few successfull
calls (either reading or writing - how soon and which one actually locks
seems to be pretty much random) when used with the new scheduler.
Downgrading 2.2.9 to use the old 2.2.7 scheduler works, raid1 mirrors are
cleanly synched. I downgraded these files to 2.2.7 (by reverting the
patches from patch-2.2.8)
arch/i386/kernel/process.c
arch/i386/kernel/entry.S
include/asm-i386/system.h
include/linux/sched.h
kernel/sched.c
This probably isn't a good idea for a production system, but it might melp
someone who actually knows what he's doing to figure out how to really fix
the problem. :-)
Bye, Martin
"you have moved your mouse, please reboot to make this change take effect"
--------------------------------------------------
Martin Bene vox: +43-664-3251047
simon media fax: +43-316-813824-6
Andreas-Hofer-Platz 9 e-mail: [EMAIL PROTECTED]
8010 Graz, Austria
--------------------------------------------------
finger [EMAIL PROTECTED] for PGP public key
