Terminal Servers on ESXi

[kullnd]

All,

I was wondering if I could get a quick bit of advice on something that I am about to setup, any comments would be appreciated:

I am planning to run a total of 6 to 8 virtual Windows Server 2008 Terminal Servers 32bit across two Dell PowerEdge R710 servers. These are stand alone servers and will not be configured for HA.

Dell Server Config is dual Xeon E5620 Procs (8 cores), 2.4GHz, 16GB of RAM

Both servers will run local storage (no SAN), with the VM files running on (6) 15k SAS drives (seperate drives for VMWare install)

Is the best RAID configuration for this type of server going to be putting those 6 drives in RAID 10, or should I split them up --- Disk IO is a big concern for me based on past experiences with VMWare --- Keep in mind that dropping one of my terminal server instances is NOT critical as there will be 5 - 7 virtual servers that can pick up the slack.. Actual design requres me to keep 3 - 4 of these up and running at any one time across the two servers.

How many vProcessors should I put on each terminal server, (thinking 2?) I have read some things that have indicated that multiple processors on the VMs can cause issues?

Thanks,

Nate

[idle-jam]

For Disk IO, please make sure the disks are connected to a RAID card with battery back write cache. For CPU, i will start with 1 vCPU and gradually increase if needed to.

Is there any estimated sizing? or an initial TS server? you can run a perfmon to look into the cpu/memory/disk before and then only deciding what's best for the virtual.

iDLE-jAM | VCP 2, VCP 3 & VCP 4

If you found this or any other answer useful please consider the use of the Helpful or correct buttons to award points.

[mittim12]

The co-scheduling of multiple vCPU is more efficient in vSphere so 2 vCPU isn't going to be that bad. I would definitely start with 1 and only increase if needed. You can read more on the cpu scheduler here, http://www.vmware.com/files/pdf/perf-vsphere-cpu_scheduler.pdf.

As far as storage goes it would be a great idea to do some IO testing to determine the best storage configuration.

If you found this or any other post helpful please consider the use of the Helpful/Correct buttons to award points

Twitter: http://twitter.com/mittim12

[a_p_]

Welcome to the Community,

I set up this kind of configuration multiple times with XenApp5 on W2K3, however it should be equal for W2K8.

The configuration I usually have:

- 2 QC Intel CPUs (54xx or better)

- min. 20 GB RAM

- RAID controller with at least 512MB BBWC/BBU (battery backed cache) for write-back operation

- either 2 SAS disks 10k or better in RAID1 (my avarage virtual disk size for XenApp is ~30GB)

- or 4 disks 2 x RAID1 (see reason below)

- 2 NICs (1 vSwitch with both NICs connected to 2 different physical switches)

- up to 4 VMs on one host (1 MS Enterprise license allows you to setup up to 4 VM's on one host)

- 2 vCPUs per VM (this performs better for a XenApp workload than 1 vCPU)

- RAM per VM -> total amount of RAM minus ~2 GB (for the Hypervisor) divided by the number of VMs (e.g. with 16GB and 4 VMs --> ~3.5GB per VM)

- NO RAM OVERCOMMITMENT ! I usually set the memory reservation to the configured amount of RAM. This way I avoid memory ballooning and I also save disk space, because the swapfile does not need any disk space.

- Create the virtual disks as "eagerzeroedthick" (check the FT/Cluster option in the create disk wizard) for better disk performance.

The reason for configuring 2 x RAID1 instead of 1 RAID10 is that most RAID controller cannot split a RAID10 into multiple logical volumes. I like to keep the Hypervisor in its own logical volume, which provides the ability - in case of a desaster - to reinstall it without affecting the VMs on the VMFS volume. Therefore I split the RAID into a 10 GB logical volume (~5 GB for the Hypervisor and scratch partition plus a small VMFS datastore for e.g. ISO files) and a logical volume for the "production" datastore.

André

[kullnd]

Thank you all for your responses, --- The two options that I am looking at for disk right now are:

2 Disks in RAID 1 for the hypervisor

6 Disks in a RAID 10 (15k SAS) for the VMs

or

2 Disk in RAID 1 for hypervisor

3 sets of RAID 0 stripes for the VMs (each VM with a dedicated set of physical disks and LUN)

Seeing as they are terminal servers, in a TS cluster, the impact of loosing one is not a big deal in the event of a disk failure, the users will just log back in to a different server... I guess I'm just wondering if the performance increase would be worth the extra time involved with reloading one if a disk did fail. It sounds like there has been some improvements in disk IO on these configurations, which makes me feel a bit better ---

The bad thing is that I am not getting alot of time to play with this before it has to go live in production... One option I probably do have is to set up each of the two servers on the two configurations, see which one performs best, and then reload the second server to match the better performer.

The server load on this is going to be 50 - 60 active users upon go-live, across the 6 virtual TS servers... I do not have the option of using 64 bit and the applications being run are largely JAVA based... I have a 5 year old 2003 server running right now that is handling about 20 users before getting stressed out in our training environment for the new system, it has 4 cores 2.8Ghz and 4GB of RAM... The RAM is generally the bottleneck... All clients will be running on Wyse Thin Terminals, so they will also be doing alot of work in Office, outlook, etc. as the TS environment is their only available environment on the thin clients. Based on this workload, I am thinking 3 virtual TS per physical server, which should give me the capacity to actually loose an entire physical host without taking my people offline (possible performance hit, but still able to work)

Thanks again for everyones responses

[a_p_]

2 Disks in RAID 1 for the hypervisor

6 Disks in a RAID 10 (15k SAS) for the VMs

or

2 Disk in RAID 1 for hypervisor

3 sets of RAID 0 stripes for the VMs (each VM with a dedicated set of physical disks and LUN)

I think the "RAID 0" is a typo and you actually mean "RAID 1".

IMO 8 HDDs are probably oversized for this workload. However - as I recommended above - create a RAID 1 on the first 2 disks and split it into 2 logical volumes. With the other HDDs I would go with RAID 10 for best performance.

... it has 4 cores 2.8Ghz and 4GB of RAM... The RAM is generally the bottleneck...

That's what I usually see, too. This is why I like to setup Windows Enterprise VM's. Once you need more RAM, add more RAM to the host, assign it to the VM's, done.

With the workload you mentioned, you may want to keep an eye on the network load. Although - from my experience - I never needed/used more than 2 NICs in a host for this configuration.

André