Is my DL380 G3 overloaded?

[theaoh]

Hi

I take care of a very small business who has a DL380 G3 with the following specs:

2 x 2.2Ghz Xeon MP CPU

4GB RAM

VMs on local storage across 6 x 73GB 10k SCSI3 HDDs (RAID 5)

I have 6 VM's currently running on this machine. As it is only a small business most of these machines are only utilised lightly (hence low RAM allocations). The machines are

1 x Windows SBS 2003 - 2 CPU, 1.5GB

1 x Windows XP - 1 CPU, 620MB

1 x Windows XP - 1 CPU, 512MB

3 x Windows XP - 1 CPU, 256MB

I know some of you may look at the RAM I have allocated to these machines and scratch your heads but I can assure you that as most of these are just simple VMs running services like FTP with very few clients and that the VMs themselves have enough memory.

My main concern is with CPU utilisation and slow disk access considering it is RAID 5, SCSI 3.

Does anyone have any thoughts on this and any suggestions on how I can test to see if I am maxing out the disk array most importantly?? There seems to be some slowness inherit with disk reads and writes. Also, if there were occasions when a VM did use more than the allocated physical memory, would this result in a large performance hit to the entire ESX server as a result of paging to the slow disk array??

Thanks

[AndreTheGiant]

You have only a old system with few RAM and old proc (with 4 core).

Also RAID 5 can be good for read I/O but bad for write I/O.

You have to check how much RAM your system use with VM powered off (I suppote about 900MB), and own are the resources with working VMs.

Also you can try to use resource pool to have differenc priority for Server and client.

And be sure that each XP has the same version, and same SP/fix (to share some memory between XP).

IMHO (if you have a single ESX) you can try to use processor affinity to block your SBS on a CPU, and share the secondo CPU between XP VM and service console.

Andre

**if you found this or any other answer useful please consider allocating points for helpful or correct answers

[theaoh]

Hi

thank for your input. I have since added an extra 2GB of RAM to the server to bring it up to its maximum of 6GB (max using 400Mhz bus CPU). This has allowed me to allocate more RAM to each VM (haven't changed around ESX allocation, swap etc). I have also removed an XP VM and swapped two XP Pro desktops for one W2K3 Server VM as it allows two simultaneous RDP log ins. The current guests are now:

1 x W2K3 SBS with 2GB of RAM

2 x XP Pro with 1GB RAM each

1 x W2K3 Standard Server with 1GB RAM

I still find however hard disk performace to be very slow. For instance, running a defrage on the 60GB virtual disk on the SBS server took over an hour to get only 5% through. I am using the onboard HP 5i RAID controller in RAID 5 as mentioned. It doesn't have battery backed cache and I had read recently that this can cause the controller to actually disable its cache, further reducing performance. Anyone familiar with this?

I was thinking, to try and improve the disk performance I would change the RAID 5 array to RAID10, striping 3 of the disks and then mirroring these disks with the other 3 disks. Now I have less VM's on the server, I should have enough space in RAID10 to do this. I was also thinking I might upgrade the from the onboard 5i controller with the small (possibly disabled) 64MB of cache to a HP 6404/256MB RAID controller with Battery Backed cache and 256MB of cache. I was thinking the combination of faster RAID and the new different controller with 256MB of cache could help.

What are peoples thoughts on migrating to RAID10 from 5 and changing the RAID controller from the 5i to the 6404 with 256MB and BBWC? Also, what is a good way to benchmark the performance before and after these changes to determine the results of these changes?

Thanks

[AndreTheGiant]

I suggest to use a controller with a battery to be able to enable write back cache policy (on RAID controller).

Then verify again disk speed.

And only if is still low change the RAID to 1+0 level.

Andre

[theaoh]

Thanks again for the suggestion Andre. Can you suggest a benchmarking method to use before and after the controller swap to determine what performance increase was achieved? Also, will there be an issue with booting ESX after changing the controller that you know of? Should I install the new controller while the hard drives are still attached to the 5i, let it boot so the driver installs, and then shutdown and connect the drives to the new controller? Haven't done this with ESX before.

Thanks

Simon

[AndreTheGiant]

You can use several tools.

But if usual use the simple dd command, like this:

cd /vmfs/volumes/YourDatastoreName/

dd if=/dev/zero of=test-delme bs=64K count=2048

rm -f test-delme

Andre