Max VM's for CPU

[KingThong]

Hi,

I have an Intel Core i5-2400 with 8GB Ram, SSD, Gigabyte GA-H61MA-D2V with VT and VT-d etc and I am running VMWare Desktop 8

I'd like to upgrade to 16GB Ram, however I'm not sure if its worth the investment if I can't run more VM's to utilise the extra Ram.

Am I restricted to running on 4 VM's using all 4 cores, or could I theoretically have 8 machines running, each with 2GB ram and each with one vCPU?

Just looking for a definitive answer really as to whether I can overcommit CPU and if it would balance the CPU load for all those VM's across the 4 cores?

Thanks for any help anyone can give

[continuum]

You can overcommit virtual RAM or vCPUs until the doctor comes - so you could assign a total of 50 Gb of vRAM or a total of 50 vCPUs as an example

However performance will suffer if you go beyond a reasonable value.
Where the reasonable limit is for your type of usage you will need to find out yourself

[KingThong]

Hi Continuum,

Thanks for replying.

I've seen poor performance before with RAM as once it starts paging it suffers.

So the limit for vCPU isn't directly corrolated to the number of Cores your host has? Is there a recommended/approved ratio (i.e. 5:1 etc) per core?

Would I be better running one VM in Desktop with ESXI 5 and assigning all the resources to that?

[EdWilts · Accepted solution]

If you have an SSD, you can page/swap to SSD instead of hard disk and that will improve your performance.   It should be a lot less expensive to add a 120GB SSD to your workstation than to add a bunch more memory and depending on what you're using it for, this might suffice.  We don't always NEED maximum performance.

The core limit is simply that you can not have a single VM with more than you have cores in the host.  i.e. if you have 4 cores, no VM can have more than 4 vCPUs configured.

I've seen oversubscription of 5:1 without any impact, but as always, IT DEPENDS.  You shouldn't configure vCPUs that you won't use but you can if you occassionally need them.  If your VMs are cpu-bound, then oversubscribing them will cause them to wait for a physical core to free up to run.

You'll simpy have to watch it in your own environment to see how busy they are.  You can monitor the CPU READY times - that's available in esxtop and in vCenter Server.

[KingThong]

Thank you. I hadn't considered SSD paging to be honest. Although the cost of Ram V SSD is actually the same price for my machine.

As long as I can over subscribe CPU then I can run more machines. Which is very helpful.

Thank you for your advice and help

[continuum]

note that Eds mentioned ratio of 5:1 can be seen on ESXi - which does this way better