Accepted Solutions
how does ESX actually go about sharing the Quad Core CPU among the VMs assuming all the VMs are CPU intensive, does it allocate each VM a specific time slice or?
Same way Linux / Windows treats processes for threading, time slice. ESX works the same way, it treats VMs like processes, so it threads them across all the CPU according to priority.
how does ESX actually go about sharing the Quad Core CPU among the VMs assuming all the VMs are CPU intensive, does it allocate each VM a specific time slice or?
Same way Linux / Windows treats processes for threading, time slice. ESX works the same way, it treats VMs like processes, so it threads them across all the CPU according to priority.
how does ESX actually go about sharing the Quad Core CPU among the VMs assuming all the VMs are CPU intensive, does it allocate each VM a specific time slice or?
This is done via the hypervisor, it handles the load balance, which as Weinstein said giving VM's multiple CPU is a bad idea, you are causing more overhead for the vmkernel / hypervisor to manage, which increase ESX load, which takes away CPU from other VM's, which causes performance issues on every VM.
Additionally presuming if I allocated the VMs by CPU only without any reservations or limits when it does the time slicing if I have a 2CPU VM it will hold the 2Cores on the physical CPU for a given amount of time, then release it and another VM will get a time slice of a CPU, is that correct?
This is all done at billions of times per second. It's not any different than services / processes running on your computer right now. If you have dual core, how does Windows manage those processes? Same principle for ESX, it figures out which CPU is free (or relatively less load) and allows that VM / vCPU to run for that given time slice on that CPU, and it cycles among the VM's on the host. Both vCPU for a VM are not scheduled on the same physical core, they are kept separate, which as Weinstein explained, means that if you don't have enough free CPU resource your VM waits until it has the resource to run.
But what if I put limts/reservations on the VM how does that affect the time slicing of the CPU time to the VMs.
Reservations just means at the time you run a VM it has guaranteed resource, setting priority only affects the running VM's when the ESX server is busy. If it's not working very hard ALL the VMs are time sliced equally, when it get's above 90% utlized, then it will be forced to decide which VM's get the resource, which is when your share / priority come in.
Lastly when I look at the summary for a VM I see the amount of CPU used how does that equate, if I've allocated 2 CPUs (physical server eg. Quad Core 2.933GHz (4*2.933GHz)) 6000GHz but its only using 2000Ghz(not heavily loaded VM), does that mean when the VM is given its slice of the CPU time it only actually utilizes eg. 2000GHz of the CPU processing power avaible on the CPU' s (2 * 2.933GHz, approx 6000GHz).
Yes the stats are from the host side, it's giving you the perspective of what impact it has for host resources to give a VM its allocation of CPU cycles. So if it's running 2000ghz out of a possible 6000, that's less than a third. You should size your VM's according to overall performance needs NOT peak. so if you have an app that runs for like an hour a day and it may use up that CPU, the other 23 hours its idle, not a good use of resource. Give it only 1 CPU. You should only give VM that consistently use CPU for more than 80% of the time, but then those are probably not good candidates for Virtual anyway, so it's a self fulfilling prophecy.