FC4 almost totally unresponsive under heavy disk IO...

[n0xlf]

Here's my config, running FC4 x86_64 SMP:

ABIT SU-2S motherboard
Dual Opteron 246
2 gigs ECC RAM
2 Maxtor 300 gig Maxline IIIs in RAID-0
Onboard Marvell 88SX6081 SATA-II controller

Whenever there is any heavy disk I/O (such as copying/uncompressing large files), everything else comes to a screeching halt...Dovecot connections won't even respond at all...Top shows no unusual processor loads, so it must be completely clogging up the bus somehow...

Disk performance shows as:
/dev/md0:
Timing cached reads: 3808 MB in 2.00 seconds = 1903.88 MB/sec
Timing buffered disk reads: 372 MB in 3.00 seconds = 123.83 MB/sec

My guess is that this has something to do with the Marvell driver, which is proprietary. The controller is not supported under the kernel/libata...

Am I on the right track here? I can't think of anything else that would be causing this...Luckily I rarely see the kinds of disk loads that cause this, but even so, it renders my server useless when I do...

[SlowJet]

Seems normal processing to me.
There is no seperation in the type of disk i/o (seperate disks and no sepration in process scheduling priorty. Everything run as fast as it can and is treated on as FIFO.

Try seperating the mail from file server disks.
Look up the Linux nice values (scheduling) and give the heavy i/o tasks a higher value (I think in the 60's) so 67-69?

SJ

[n0xlf]

Wouldn't that only change the scheduling of the process(es) themselves, and not affect the bus side of things, which is where I think the problem is? During these heavy disk I/O times, I might see 50% processor usage on one processor while the other processor is completely idle...

I see what you are saying about there being no priority in disk I/O, so a simple mail check is actually fighting for a slice of 100% utilized disks...Hummm...