Ivy Bridge CPU and host performance charts

[gadvdi]

Since I've been using Ivy Bridge (E5-26XX v2) CPUs, I have noticed that when I look at the host CPU performance in vcenter or in the client directly connected to ESX I find that the CPU is able to exceed 100% usage, as in this graph attached.

The Host in question is running ESXi 5.5 on a Dell R720 with 2 E5-2690 v2 CPUs, for a total of 20 cores each running at 3.0 GHz.  Theoretically this would produce a maximum of 60000 MHz, but you can see in the chart that the usage reached 67000 MHz in this particular test run.   How should I read this to determine if my Host is overloaded if it can exceed 100%?   What is the real max CPU output with Ivy Bridge?  I originally saw this on ESX 5.1 and assumed it would be corrected in later updates, but if ESXi 5.5 still shows the same pattern, I can't assume that a rational reading of CPU performance will be forthcoming.

[ldesfontaines]

Hi,

I don't see any problems here. E5-2690v2 CPU is a 10cores @ 3GHz CPU and is also "Hyperthreading enabled".

I do think that, in your setup, you have 20 cores @ 3GHz, but finally 40 logical CPUs (ie threads) seen by ESXi.

In this case, you may have more than 60GHz available in this host. So, no problems with your chart.

Hope it helps.

Ludovic

[gadvdi]

Thanks for the reply.  If the question is "Is this host fully utilized?" then what is the answer?  If 100% does not mean fully utilized then what does?  70000 MHz? 80000 MHz? 120000 MHz?   Do i need to just push it until something breaks?   The chart should reflect a useful metric.  110% is not a useful metric.

[Linjo]

I think it has to do with the turbo-boost functionality of the CPU, the processor have a rated frequency that it advertise (lets say 100) but in some cases it can go over that speed (110) and then the graph will be skewed.

See more about turbo mode here: Intel Turbo Boost - Wikipedia, the free encyclopedia

You could probably disable this in the bios to see if it is related.

Would that explain what you are seeing?

// Linjo

[ldesfontaines]

Hi gadvdi,

I still think that this is HT related. In fact, when HT is enabled, the global available power is still based on the number of socket & the number of cores per socket.

HT is not used to calculate the available power.

In you case, you still have 2 CPU x 10 cores x 3GHz = 60GHz, but, HT enables the ESXi to overload a little bit this "power". This is why you have a "used" CPU Capacity graph able to draw at more than 100%.

To confirm, check your ESXi summary page: It must show in the General Section :

• Processor sockets : 2
• Cores per socket : 10
• Logical Processor : 40
• Hyperthreading : Active

And it mus show in the Resources Section :

• CPU Capacity : 20x 3GHz

Hi Lingo,

I don't think Turbo Boost is the answer. When an Intel CPUs disables some cores to activate Turbo Boost, the Global power is reduced. For example (these figures are not the right, but are used to ilustrate) :

E5-2690v2 : 10 cores @ 3GHz, but when Turbo Boost is activate, maybe 6 cores are available @ 3,5GHz, so a global CPU capacity availability of 21GHz instead of 30GHz in the first case. To be able to have more CPU Capacity with a 6 core Turbo Boost enabled CPU, its frequency should be 5GHz, which is not available at the moment in "enterprise class" CPUs.

Hope it helps.

Ludovic