Cisco ISE CPU Allocation?

[mhost]

Hello,

We are in the process of implementing Cisco ISE - and have received the "hardware requirements" for installation in VMWare.

I am a bit puzzled by the default CPU configuration in the Cisco ISE OVA template.
It seems that 16vCPUs/Cores are allocated to each VM - but at the same time there's a reservation/limit of 16000 MHz.

Can anyone explain how this affects CPU scheduling on the hosts?

The way I see it, the VM is limited to a CPU capacity WAY below the allocated CPU capacity - and will never benefit from the 16vCPUs?
But how much does this "disturb" the CPU scheduling on the hosts?

PS: We have 2x Xeon 6154 in each host. (2x 18 physical Cores @ 3GHz)

Best regards
Martin Holst

[vmrale]

Hi,

beautiful. :-)

16 cores per socket - maybe because this application benefits from shared cache between cores on one socket, but I'm not brave enough to explain the rest.

Contact a Cisco representative and ask them for some explanations.

[mhost]

Thank you,

Our current (6 vCPU) ISE test appliances seem to spread the load over all available cores - no matter how low the CPU usage is.
(Whereas our SCCM server nicely "parks" unused CPU cores).

Someone else suggested that the billion-core approach might result in lower latency for the ISE appliances.

But I fear that the constant noise on so many cores may result in high ready / latency for other VMs.

/Martin

[daphnissov]

Virtual machines do not operate like physical machines and throwing more cores/CPUs at VMs has the opposite effect of increasing performance or decreasing latency. It does the exact opposite, in fact. That's why I advised to monitor the utilization of this VM to determine if it's oversized. If it is, you will actually gain performance and responsiveness by decreasing allocated vCPUs. This subject is written on extensively and, if interested, I can supply some resources to read more in-depth.

[daphnissov]

It's probably a bug as there's no reason to place a limit on a vendor appliance. The correlation between 16 sockets and 16,000 MHz seems to suggest someone didn't know what they were doing when they authored this appliance. I'm not familiar with the ISE but I know Cisco usually take a hard-lined approach to their appliances and will refuse to support customers if they've jacked with them. In the interim, I'd remove the limit and reservation but keep the 16 vCPUs as is. If those vCPUs aren't fully used they're going to cause performance degradation for the rest of your VMs in a shared environment, so that is of more concern once it's in operation.