Hey Ex,
That million dollar question has been asked many times before. The simple answer - there is no definate answer. There are so many variables that contribute to the # of VMs you can run on a sever:
- # of Sockets
- Core per socket
- CPU speed
- Memory density
- Network bandwidth Allocated per Host
- Storage connectivity & bandwidth
- VM attributes & workload function (amount of memory, CPU & disk per VM)
Even though you can probably squeeze a much higher # of VMs per server with increased memory densities & CPU power, also increases your failure domain. HA might be able to "recover" VMs on a failed host, but how many VMs are you willing to lose in one go? When sizing your hosts, also take into account failover capacities. Each of your host clusters should be able to tolerate at LEAST one host failure and meet all the memory & cpu requirements of the extra VMs now spread accross remaining hosts in the cluster. For this reason adminstrators don't often push past 60% capacity of their hosts. VMware uses a "slot" size calculation to ensure you have adequate failover capacity in your clusters.
As a typical metric I regularly see around the 20 VM mark per server on a typical 2 socket server with 96GB memory and dual 10G uplinks. It's also good to consider that most software licensing (VMware, Nexus 1000v etc) is based on socket counts, and since memory is going to be your first contention point - servers that can support high amounts of memory are usually an ideal choice. B250 is perfectly positioned for virtual workloads like this, while the B230 is a little firecracker with a little more CPU power under it's hood if you need it.
As vague as I've been, hopefully this starts your thought process
Regards,
Robert
