Re: Strange disk performance issue
On Tue, 2011-12-13 at 13:23 -0600, Stan Hoeppner wrote:
> > hoping for help, I'm experiencing a strange performance issue on debian
> > squeeze. The host is running 2.6.32-5-amd64 on a Xen enterprise cluster,
> > filesystem lies on a NetApp storage mounted using ISCSI. Other virtual
> > hosts on that cluster does not have the same performance problems.
I have to correct that statement. All hosts have performance problems.
> > As soon as I start to copy i.e. a 2GB file local or to a remote host,
> > the iowait is rising and the host feels jerking.
> >
> > sar reports %iowait > 60
> >
> > Activating block_dump for a minute (echo 1 > /proc/sys/vm/block_dump)
> > shows me that the process causing the iowait is jbd2 which is part of
> > the ext4 filesystem.
> > Any ideas howto tweak the fs for better performance?
>
> Ext4 isn't the source of the problem but simply the messenger in this
> case. Try the deadline and noop elevators and see if that helps. The
No it doesn't, same bad performance regardless of which elevator is
used.
> default CFQ elevator is known to suck. Also, post the block layer
> device driver you're using to access the virtual disk.
xen-blkfront is loaded.
Regdards
Frank
Reply to: