performance on ESXi

[Pille66]

Hello all

i try to get some information about performance measurement of an ESXI server.

in the performance tab of the esxi are a couple of counters - for example CPU, Memory, Storage ...

what i have experienced and read - if the CPU ready time is hight the system leaks on performance

we had over 35000 which is 175% rdy.

i reduced a problem_vm from 4 to two processors and the ready time is down too 300.

my questions:

which counters indicates a bottleneck ?

how the ready time can be explained ?

any good staff for performance tuning of an esx server ?

thanks all

[weinstein5]

Ready Time is the measure of time that VM is waiting for its vCPU to be scheduled to a physical CPU - so I high ready time is an idication that there is contention for CPU cycles.

This link should help you our - http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.1.pdf

[Pille66]

hello

thanks, i know the document

but they is no answer to the questions

let me express it on a different way.

some people belive, because it is possible, to store on a 4 CPU Server with 10 cores (which is 40 vCPU) round about 50 VMs. every VM have more as 2 vCPU. some of them have up to 8 vCPU - i see a overcommitment of CPU/vCPU - isnt it ?

Also its a problem to store SQL Server with 8 vCPUs and 10 "small" VMs - why ?

now i want to know where exactly the bottleneck is

which VM pull the Datacenter down (some VMs are just doing fine others are slow on performance and a high ready time) - why ?

i want to know which performance counter shows that the ESX is "out_of_resources"

how can i influence the performance ?

we have DSR installed

i cant find any guide which explains the counters

i cant find any guide with calculate the max possible VMs, performance, ...

mfg

Juergen

thanks

[Pille66]

no further suggestion ?

more questions

on the host is written

processor sockets 4

cores per socket 10

logical processors 80

why 80 ?

i want to get a technical explaination why reducing of 2 CPUs (from 4 to 2) in increase the performance multiplyer 2

thanks

[PhillyDubs · Accepted solution]

There's no way to have a magic calculation to determine how many VM's fit on a host as workloads are rarely ever perfectly consistant nor is hardware all the same across every environment.

With that said, you  have found your bottleneck, CPU. It is typically seen as best practice to start with 1vCPU for your virtual machines and scale up if more vCPU is needed. Start by reducing the number of vCPU on virtual machines, especially ones that don't even peak above 60% total usage.

Also, to find which VM is consuming your CPU, just observe which virtual machines are consistently using the most CPU and the highest amount of vCPU. Those virtual machines will typically be the ones that have high CPU Ready % and will cause an impact to other virtual machines that want CPU resources.

Someone in this thread has already explain what CPU Ready time is, I'm not sure what else you want to know about it. It is typically the counter I use to see if my vCPU is overprovisioned though.

[Pille66]

hello Philly

thank you for your answer but to be honest it doesnt help much

no offence - but i need an official document if i am telling the company to put 50k for a new server on the table.

i need to explain why, in the example earlier, reducing the vCPU from 4 to 2 increases the performance dramatically (ready time from 35000 to 2000-3000). i need to explain (on an technical level and non technical level) that the server is overprovisioned. i need to explain why i recommend an new server. i need to explain why its not a good move to put productive SQL and Dev machines on 1 Host. 1 need to explain the hardware resource management. i want to know how i can figure out an 'possible' bottleneck on the storage (how i can recognize it)

i have read already that its better to have one vCPU instead of 2 - and you spotted it too. but what is the explaination for it (official). if i want to change the standart from 2 vCPU to 1 vCPU i need to convince a lot of people. i need something official and not just some forum entries - no offence please

From my understanding, and it looks quite logic, if the Hardware have 4 CPUs and 10 cores that max 40 vCPUs can be configured and not 200.

it should be possible to find performance Docus, performance counter docus.

Sorry, but opposite of mine are standing 10 people which all telling me that i am wrong (even i proved it with the ready time stats before the change and after). but they dont understand and me as well.

best regards

[PhillyDubs]

VMware has all this information for you.

http://www.vmware.com/resources/techresources/

There are also hundreds of blogs, both by VMware(  http://blogs.vmware.com/ ) and individuals that will provide you official and non-official resources to answer your questions.

This link should provide you the VMware white papers on performance: http://www.vmware.com/resources/techresources/cat/91,203,96

Begin by reading those and with that information you should understand the why and how that you are desiring.

[PhillyDubs]

80 logical processors as you are using Hypterthreading. If you turn Hyperthreading off, you will only have 40 logical processors.