[rhelv6-list] Performance issues with RHEL6 (iowait)

[Matthias Saou]

William Warren wrote :

> On 2/9/2011 4:47 AM, Matthias Saou wrote:
> > Hi,
> >
> > I just got this isolated RHEL6 server, with hardware I've never used
> > before. It's a Supermicro X8STi motherboard (Xeon i7 920) with two cheap
> > SATA disks (WD10EADS-00P8B0) using software RAID-1.
> >
> > The system is unusable : A quick strace of various processes shows very
> > long waits during many different I/O related calls such as open, read,
> > pread, stat, fdatasync...
> >
> > I've tried failing every other partition in order to be sure to use only
> > one disk for a while, with the same results. Running RAID rebuilds
> > sometimes peaks at 60-100MB/s but very rarely. Right now for instance,
> > the main 900GB RAID-1 is rebuilding from sda2 to sdb2 and it's
> > reporting 400kB/s with system load of over 2 (the server is idle apart
> > from this rebuild). While this is going on, even smartctl on sdb takes
> > time to respond. When no rebuild is in progress, the problems are the
> > same.
> >
> > I'm using the latest released 2.6.32-71.14.1.el6.x86_64 kernel, and
> > here is the detail of the SATA controller(s) :
> >
> > 00:1f.2 IDE interface: Intel Corporation 82801JI (ICH10 Family) 4 port
> > SATA IDE Controller #1 Subsystem: Super Micro Computer Inc Device 0009
> > 	Kernel driver in use: ata_piix
> > 	Kernel modules: ata_generic, pata_acpi, ata_piix
> > 00:1f.5 IDE interface: Intel Corporation 82801JI (ICH10 Family) 2 port
> > SATA IDE Controller #2 Subsystem: Super Micro Computer Inc Device 0009
> > 	Kernel driver in use: ata_piix
> > 	Kernel modules: ata_generic, pata_acpi, ata_piix
> >
> > No errors reported by the kernel apart from some processes stuck for
> > more than 120s before the last reboot. Has anyone seen disk I/O issues
> > like this with RHEL6 and similar hardware?
> >
> > Matthias
> >
> > PS: A possibly important detail is that this is a server provided by
> > ovh.com which has been offering RHEL6 for a while but still has it
> > marked as "BETA" (it's not the RHEL6 beta, it's how they consider their
> > offering of RHEL6), which might indicate known problems between their
> > typical hardware and RHEL6. I didn't find any clues with a quick
> > bugzilla search...
> >
> are you using the bios raid or MD raid?  If you are using biod raid you 
> are running everything effectively pio mode and it's going to suck.  Put 
> the bios in ahci(not raid mode) and then use Linux raid from inside the 
> partitioning scheme(which is md raid) i bet performance will be much better.

I probably forgot to detail, but it's indeed MD RAID I'm using. No one
in their right mind would use one of those cheap motherboard pseudo
RAID, would they? :-)

Anyway, I've reinstalled the server with RHEL5 and it's running fine
now. Definitely something going on between this particular hardware and
RHEL6.0... since I intend to use it only as a KVM hypervisor, and that
RHEL6 still doesn't support the new cool stuff that would make a
difference (like KSM), I'll stick with RHEL5, even though all of the
virtual servers running under it will be RHEL6.

Matthias

-- 
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Fedora release 14 (Laughlin) - Linux kernel 2.6.35.10-72.fc14.x86_64
Load : 0.07 0.07 0.08