Exchange 2003 - Poor performance

[JeanB]

Hi,

We have set 2 Vmware servers running ESX 3. On one server with have Exchange 2003 and a Windows 2003 DC. On the other one we have a file server and we are planning to install SQL 2005

The hardware setup for both servers is:

HP DL380 G5

(2) Dual-Core Intel Xeon 5160 Processors (3.00 GHz, 1333 FSB)

4GB of memory

2 internal disk on 72gb

Both servers are attached to a HP MSA 500 device. The device has 7 x 300gb 10K disks Raid 10

Actually the disk space is seen as a big disk by VmWare.

We migrate our first 10 Exchange users and we already have performance issues. Exchange is very slow to respond to Outlook users.

Exchange is configured as recommended by Microsoft.

The latency created by VmWare seems to be the problem.

Where I should looking or to solve my problem....

Thanks

[esiebert7625]

You might give this a read...

Deploying Microsoft Exchange in VMware Infrastructure - http://www.vmware.com/pdf/exchange_best_practices.pdf

[mrcreosote]

I have experienced a similar problem in the past. In our case it was caused by CPU scheduling problems - we had configured the exchange VM to use 2-VCPU's on a host already laden with other VM's.

Monitoring the 'CPU ready' counter will help determine if this is the case.

[JeanB]

Thanks for the response

I have only two others VM with very low activity on this machine.....

[mpverr]

For grins - get VizionCore ESXCharter (free version). That will give you some good metrics on the systems.

Have you run perfmon within the Exchange box? are you seeing Disk queing taking place?

have you monitored /var/log/vmkernerl and vmkwarning?

Is this 3.01 all patches?

[gbowers]

Try logging into your console directly or via ssh, launch "esxtop", expand the default size of the window and look for a column named "%rdy". You should see the running processes and guests, if your exchange server is above single digits than you are running into a scheduling issue. this can be caused by high CPU on the guest or an overly aggressive number of multicpu instances.

The way ESX handles CPU requests is to schedule all threads on a per instance basis, simultaneously. Which is to say that if you had 4 cores and 8 2cpu guests only two of those guests will receive cpu resources at a time.

[SouthEast-Smith]

10 Exchange VMs with on a host with only 4 GB of memory? Did I read that correctly.

Things to improve performance

\- Put the VMs on different LUNs so they are not using all of the same disks

\- If DRS is enabled, you don't have to worry about balancing out the machines based on CPU

[SouthEast-Smith]

Sorry, I see 10 users, not 10 servers :S

I wouldn't go for the one huge LUN spanning all of those disks. Use multiple smaller LUNS and put page on one C: on another and OS on third - You will be using more disk of the SAN that method

[MattG]

It sounds like you are having either a disk or network issue or VM resource contention issues.

Like someone said earlier check your PerfMon "CurrentDiskQueue" sizes. Another Perfmon Disk counter to check is Average Disk Sec/Read & Average Disk Sec/Write. This number should always be below 50ms.

Also check MSExchangeIS/RPC Requests object in Perfmon:

http://technet.microsoft.com/en-us/library/bb124154.aspx

If you plan to migrate high IO apps to ESX, you want to eliminate VM resource contention issues. For example, you can over subscribe memory and CPU. This may work fine with smaller servers, but it will definitely increase your virtualization "penalty" with high IO apps.

-MattG

[Anders]

You are not running a single MSA 500 as a shared device do you?

We dont support shared SCSI anymore.

If not you might have a network duplex mismatch or a HW error.

I've also seen exchange perform poorly with several mis configurations in AD.

Decommissioned GC or exchange servers that are not properly removed for instance.

I'll gurantee you there is something wrong, should easily get several hundred users on that...

\- Anders