Re: Exchange 2003 Disk Queues

On Mon, 18 Dec 2006 07:11:01 -0800, Darrel
<Darrel@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:

Hi,

I am looking for some information on checking the disk queues on our
Exchange 2003 cluster running on a EMC CX500 SAN.

My initial thoughts are that it is running a high disk queue, but this is
based on watching the average counters for all drives that have exchange
stores on them.

I'd get the stats off the EMC software, they're better than Windows.


The cluster has two active nodes and one passive node, each of the active
nodes has 3 drives which hold the 3 exchange storage groups.

The RAID5 array that has the stores residing on them has 4 drives and this
has two 6 luns assigned 3 for each node which in turn are allocated to the
two active cluster nodes.

I believe that a dedicated RAID5 array should of been created for each node
to store the exhange stores on rather than have contention across a single
RAID5 array.

Does anybody have any guiding thoughts on this and how to actually check if
the queue lengths are too high.

Thanks

Ignore the node aspect here.
You should create a disk set for each storage group and spread the
LUNS over them. A fresh LUN for each store. In addition you'll want
another LUN on a separate set of spindles (RAID 1 or 5) for the
transaction logs. Be careful how the Clariion software carves the
disks up though, it's none too clever about how it does it and you can
easily end up with an awful bottleneck.

So you're looking at six RIAD5 arrays each of at least six disks for
the six SG's to get the spindle count up. Use smaller disks and more
of them rather than large disks.
You'll then need six RAID1 array's for the Transaction Logs and
another RAID1 array for the SMTP queues.

You'll need another RAID1 for the quorum and DTS (obviously you've
already got this)

Then of course you can sit back and wonder why you didn't buy NetApp
so that you could just split the aggregate into a load of volumes and
have dozens of disks (spindles) giving massively better performance
than your Clariion.
.

Relevant Pages

  • Re: Exchange 2007 - Multiple SG per LUN?
    ... 1+0 LUN for max performance and also have the transaction logs (RAID1) ... Now why would combining some of these stores on one LUN be beneficial ... especially seeing as the disks have already been bought!? ...
    (microsoft.public.exchange.admin)
  • Re: Hardware Recommendations for Exchange FE/BE
    ... >>We have a small office, currently 120 employees, and we are planning on ... >>We are laying out an idea of moving our Exchange services to a colocation ... > On the BE you might will want to use six disks in three RAID1 sets. ... the Exchange Logs and finally the Exchange Stores ...
    (microsoft.public.exchange.setup)
  • Query: force-mount a corrupt UFS FS possible w/o fsck completion ?
    ... the purpose of attempting to recover *some* small portion of data on the ... -failed blocks on too many disks in a raid5 array. ...
    (SunManagers)
  • Re: 2.6.23.1: mdadm/raid5 hung/d-state
    ... We're stuck trying to build a disk server with several disks ... in a raid5 array, and the rsync from the old machine stops writing to ... lock without rsync, using 8 simultaneous dd's to the array. ...
    (Linux-Kernel)
  • Expanding partition in 3510
    ... the logical drive in the RAID5 array. ... The sccli 'show partitions' command now sees 2 partitions, ... created by the addition of the 2 disks. ... SOUTH AFRICA: Building 3, Parc du Cap, Mispel Road, Bellville, 7535, South ...
    (SunManagers)
Loading