VMware Cloud Community
vitasconsulting
Contributor
Contributor

Intel CPU C-State question...

Hi,

What do you think about C-State and ESX? Is that option should be activated on ESX or Not? By default on Nehalem Dell & HP servers that option is activated.

Thanks

Tags (3)
0 Kudos
5 Replies
RParker
Immortal
Immortal

Well in the firs place, other than enabling VT Extensions you should leave the BIOS settings at default.

And in the second place, this is C2 which means power consumption and cooling. ESX would not be affected by this setting one way or the other, but I would leave it alone.

0 Kudos
vitasconsulting
Contributor
Contributor

As I know DPM feature of vSphere is depend on P-State and C-state. So ESX4 is C-state aware software. And C-State in some case must decrease the performance due to time of wake up of processor's core (in the world of real server). Is there a recommendation or best practice from VMWARE about this feature?

0 Kudos
JoJoGabor
Expert
Expert

I have run benchmarks for VDI loads with HP Proliant G7 Servers and I got significantly more density by setting no C-states and the "Maximum Performance" power profile, both in the BIOS. I am talking nearly 40% more sessions without performance degradation. I havent tested this on the G6 servers though.

0 Kudos
AureusStone
Expert
Expert

Just to confirm.  For your benchmarks, are these the two settings you are comparing;

1) Set server to maximum performance.  Which uses no power saving.

2) Set server to allow the OS to control power management and then set Power.CpuPolicy in advanced settings to "Dynamic"

Is that right?

I was thinking of changing the hosts in my environment to setting 2.  From what I was told vSphere is able to manage the performance quite well and using this setting shouldn't drop the performance at all.

A 40% drop is quite surprising.  That would have to be a bug.

0 Kudos
JoJoGabor
Expert
Expert

I have set the power profile to max performance and set processor options to "No C States"

I hadnt seen the VMware option you mention, I was advised by vmware just to do the first two, but that looks like it could help when using the HP Dynamic power mode in tandem. I will try ans rerun a benchmark using that

Sent from my iPad, please excuse any tpyos (sic)

0 Kudos