Showing results for 
Search instead for 
Did you mean: 

EMC / ACE / AAE virtual Appliance scaling

New Contributor III
8.x release notes describe the amount of CPUs, Memory and Disk size which is needed for small, medium or large installation of Virtual Management Center, virtual Access Control Engines and virtual Analytics Engine. In Download Center are corresponding small, medium and large-OVAs.

Can I change between medium and large just by changing CPU and memory amount afterwards or are some application/cache/database parameters different in those OVAs?

Example 1:
Currently I have 2 NACs for about 10-12.000 wifi-Devices. The clients re-authenticate VERY often because everytime they roam from one AP to another, our 3rd-party-Wifi-Controller generate new RADIUS-requests. (NAC statistics says: about 2.500 auth requests per minute and 350 End-System Events per Minute for each NAC.) Those NACs are vNACs at version 7.1.2 with 2vCPUs and 12 GB RAM. Their cpu load is between 5 and 10%.

Now 8.x release notes say I should use a medium 16vCPU vNAC for 6.000 End-Systems. Problem is: In ESXi (and perhaps other hypervisors too) every time a 16 vCPU VM does something, 16 logical hypervisor CPUs are blocked for that cycle. F.e. if I have a hypervisor with 16 logical CPUs and 2 VMs with 16vCPUs, both VMs cannot work at the same cycle and performance problems would appear under load even if the load is a single thread. The %READY value rises (VM was ready but could not be scheduled to run on a physical CPU). If both VMs have 8vCPUs they could work at the same time/cycle. So it is our interest not to allocate more vCPUs than nececcary. Which leads me to my initial question: if I reduce the amounts of vCPUs afterwards (f.e. because we have lots of wifi users but dont use assessment) would there be additional changes be necessary in config files?

Example 2:
Our 7.1.2 Analytics appliance with 8vCPUs analyzes about 450k fpm and has a CPU load of about 12% with spikes up to 35%. According to 8.x release notes we should increase to 16vCPUs but that seems to constrain more hypervisor CPUs without being neccessary from performances perspective as long as no performance hungry features appear with 8.x...