vCPU versus cpu entitlement

[michaelob]

Hi,

Apologies if this is a basic question. I have to setup a number of hpvm's on rx2800 in the future I just planning the best way to utilise the cpu resource. The rx2800 will have 8 cores to share among the hpvm's. The hpvm's will be used in a functional test environment so I don't envisage a heavy workload with the hpvm's.

My question would it be better config the hpvm's with more vCPU's with lower entitlement or a single vCPU with a great percentage of cpu entitlement ?

I guess the increased number of vCPU would give a great number of run queues but a reduced time slice of CPU.

Any pointers you can provide would be much appricated

Thanks

Michael

[Stan_M]

There is no good answer except - it depends...

It depends on how many guests, what are your sizing requirements, performance expectations etc.

Generally, it makes sense to keep the guests as narrow (less vcpus) as other requirements allow.

Sharing physical cpu among many different guests does have performance impact. So maybe start with

2 vcpu guests and if needed you can always change the config - trying different config is only

 a reboot and one hpvmnodify command away.

HTH

Stan  

[michaelob]

Thanks Stan for reply,

The environment I will be building would be used for functional testing so performance initially wouldn't be a key requirement.
Following functional testing I would moved the applications on right sized hardware.

I was planning on starting with a default config for each hpvm of 1 vCPU , 4GB RAM and the default 10% entitlement. Then monitor the usage and performance and increase accordingly, so based on your replay I would be best to increase a vCPU and then the % of entitlement ?

Ideally I would like to fit as many hpvm's onto rx2800 i4 as possible. Some of the hpvm will be running HP9000 containers inside as I have to move some apps currently running on PA RISC hardware onto Integrity going forward.

How many hpvm could I run on a rx2800 i4 (8 cores) given I had enough memory to accommodate 4GB per hpvm ? What deployment config would you recommend if the goal was to stack as many hpvm's onto as few rx2800 boxes in a functional test environment ?

Thanks
Michael

[Stan_M · Accepted solution]

It really depends - for example, if you want testing on uniprocessor or MP system.

You may probably want both - both UP and MP specific bug exist. When you go to MP config,

keep it as narrow as possible (which is why I said 2 vcpus).

With minimum entitlement of 5%, you can theoretically have up to 20x8 = 160 guests.

Memory will be the limiting factor.

You can always adjust things as you go.