This question has been asked numerous times during projects or classes.
The simple answer … it depends on your environment :>
The vCenter is the central management server for all aspects of the virtual environment, from KVM functions at the guest virtual machine level, to automated patching of ESX hosts.
The server role is therefore critical to the virtual environment it is used to manage. However one important question to consider is “How long can the company tolerate a vCenter outage for”.
While the vCenter is a critical server role, the guest virtual machines will continue to function during a vCenter outage. Due to this, the majority of implementations may not require passive installations of vCenter.
During a vCenter outage, changes to the environment will not be possible via the vCenter. Therefore functionality with configuration owned by the vCenter will not be configurable, i.e. DRS and distributed switches.
Assuming that the business can withstand the loss of these types of functionality until vCenter connectivity is restored a single virtual vCenter protected by VMware HA would be perfectly adequate.
Physical vs. Virtual vCenter
|Required for Internal policy / regulatory compliance?||Fully supported by VMware|
|Dedicated resources from physical machine||Competition with other Guests for resources|
|Not impacted by a VMware host outage||Take advantage of virtualisation benefits (portability/backup/ hardware abstraction etc)|
|Bound to hardware – server lifecycle||Can be protected with VMware HA|
|Extra cost to protect for outage VMware vCenter Heartbeat|
Points to consider.
· If ESXi is used with lockdown mode enabled, ensure it is possible to access the DCUI of each host in the cluster (i.e. DRAC, ILO, etc.). Without this access, lockdown mode cannot be disabled and an engineer will not be able to log directly into the ESXi host running the vCenter to fix issues.
· If DRS is in fully automated mode and Host to vCenter communication is not functional, no guest migrations will occur. If the cluster becomes imbalanced business application performance may be affected. Thus ensure resource requirements for critical applications are satisfied (i.e. using reservations or ensure adequate cluster size).
· Disable a Virtual vCenter guest machine from automated DRS migrations. This ensures the engineer will know exactly which host to login directly into without wasting precious time searching for the vCenter.
· As VMware HA functionality continues in the event of a vCenter outage, a virtual vCenter can be protected just like any other guest machine. However, consider the restart priority of the vCenter guest virtual machine, and slot size if a reservation has been used.
Personally I would recommend a virtual vCenter for all implementations unless business requirements specify otherwise. The benefits provided by virtualisation far outweigh the luxury/advantages of having a dedicated physical machine. However there are management points to consider.