Higher writes on XFS filesystem via LVM than expected
bond.masuda at jlbond.com
Wed Sep 1 08:38:55 CDT 2010
Does iotop not show any other process or processes being responsible for
the 110Mbit/sec writes?
I would almost like to see an iostat and iotop output of the same period
to compare... something like:
iotop -n 10 -b -d 5
iostat -m 5 10
run at the same time.
Those outputs are normally consistent.... but it sounds like your
situation it is not?
Keep in mind, iostat/iotop show actual writes to disk. For example, XFS
does a lot of buffering, so one may write 500MBytes, but see the
iostat/iotop activity a few seconds later.
On Wed, 2010-09-01 at 15:17 +0200, Florian Jauernig wrote:
> Hi List,
> hopefully this list is a correct place to ask...
> I am facing a strange problem with my NFS server: although network interface
> statistics only show up about 21Mbit/s incoming traffic, I am observing
> about 110 Mbit/s writes to my storage with iostat.
> With iotop I can see, that the nfsd processes only use some thousand kbit/s.
> But total disk writes per second are about 110Mbit/s reported by iotop are
> again about 110 Mbit/s.
> I have to mention that I am no storage expert, but I tried my best with
> searching for best practises in the internet of how to configure my setup.
> My configuration:
> We have a good old Poweredge 1850 acting as a NFS server for 3 VMware ESX
> servers (Poweredge R510).
> As storage I have an EMC CX300 with 2 LUN's. Each of the LUN's is a RAID 10
> of 10 disks. The storage is directly connected (no switch, etc). Each LUN is
> owned by another Storage Processor. We have multipathd running so in case a
> fibre link is going down we have failover.
> I created one big volume with LVM that consists of these 2 LUN's.
> As filesystem I am using XFS.
> I tried my best so everything is aligned (Partitions accordingly to the
> stripe size of the LUN's), Metadata of the LVM PV's is hopefully also
> XFS is hopefully correctly tuned with sunit and swidth, and mounted with
> relatime and nobarrier.
> On kernel level I am using noop scheduler (so the CX300 storage processors
> can do the scheduling work and do not interfere with each other).
> Has anyone an idea what might cause the much higher constant writes to disk
> than expected by NFS? If you need further information like exact
> configuration or logfiles I will append those.
> Any help is greatly appreciated!
> Thanks in advance
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> Please read the FAQ at http://lists.us.dell.com/faq
More information about the Linux-PowerEdge