Provisioning vcpu's question

[ttl999]

I have a cluster of 12 host. Each host has 2 proc. sockets, 6 cores per socket, 24 Logical procs. I am running ESXi 5.0. This cluster is currently housing 225 VM's (vdi desktops presented by xendesktop) they each have 1vcpu, 3gb of mem (no reservation and unlimited recsource allocation).

My question is, I need to add 150 or more vm's for vdi to this cluster. I need to know how to determine how many vcpu's this cluster can handle so that I do not over allocate resources?

Sorry, I am new. Going to VMwtraining in 2 months but can not wait that long to resolve this.

[JCMorrissey]

Hi,

From doing the maths there your looking at 12 x 24(logical processors = vCpu's essentially) = 288 so if you have 225 vm's at present (with 1 vCPU each) that leaves you with 63 logical processor/vCPU's left before you over-subscribe.

Just so you know we have a quite a large VDI environment (running XenDesktop 5.6 / vSphere 5.0 u2) and over-subscribe to the order of 3:1 - so in essence if 300 vm's running 1 vCPU: 100 vCPU's  - and the performance is fine.

Key thing to make sure is:

i) Your storage sub-system is adequately specc'ed for IOPS

ii) You reboot idle VDI's in staggered sequences to lessen the load.

The 3:1 ratio above is a "middle ground" in our environment, for heavy duty users you might want a 2:1

Many tx

[ttl999]

Thanks JCM this will get me started in the right direction.

[TomHowarth]

In a standard environment with modern hardware, CPU contention issues are now very rare. the major bottlenecks particularly with VDI are Memory and Disk IOPs.

your current environment hosts 225 guests, even taking into account ESXi overhead you are still not in a contended state.  the addition of a futher 150 guests will take your contention rate to approximately 1-1.2 this is next to nothing.

so long as your Guests do not currently exhibit high CPU utilization you SHOULD be fine,  I say Should because I have no evidence regarding current Guest CPU utilization.  if your guest are regularly running at 90% CPU then the additional 150 desktops could be too much.

However if they are running at say 25% CPU Utilisation then you could possibly install 500 more guest on your environment without major impact.

My Gut Feeling is you will be fine. however please do your homework first by testing the resourse utilsation of your hosts and Guests over a set period, hopefully taking into account peak usage periods like Month end.

[ttl999]

Thanks, Tom. current cpu util is Average 5% on pcpu's and 1% on the vcpus. I spoke with a support and got a batter handle on which columns in esxtop to watch and monitor. After finding that we are not hardly utilizing cpu right now made me more comfortable. I am going to put some of the host in maint mode to simulate the the higher usage to test and make sure all is good.

[PhillyDubs]

I'm sure support mentioned this to you, but be mindful of CPU Ready and CPU Latency. These values will typically rise in relation to oversubscription and high CPU utilization and will have a major impact on performance for the end-user. CPU ready at 10% when viewed through ESXTOP is "typically" the point people see it as an issue.

JCMorrissey's advice is very good in relation to typical VDI performance issues.

[ttl999]

Yes, he did hit on that along with the IO conters to watch (LATrd&wr), Memory TCHD, Network DRPtx&rx. Thanks!