Cluster Capacity Recommendations

• How many hosts do you have
• How many VMs do you have
• Do you have reservations on your VMs
• Remember the limit of 32 hosts per cluster
• Do you have Oracle clustering considerations (CPU Licensing)?
• Do you have to have separate clusters because of large VMs?
• By pooling together hosts into larger clusters, DRS is far more efficient at VM placement and providing resource management. It also allows for more efficient HA policy management since the absorption of spare capacity needed for infrequent host failures is now spread out over a larger set of hosts
• Check all hosts can see shared storage
• Check your networking. Inconsistency in the network configuration may result in virtual machines losing network connectivity after redistribution, in the event they are moved to a physical host lacking the required VLAN.
• It’s critical to ensure there is complete network redundancy in all paths between hosts in the cluster. The greatest risk with VMware HA implementations is a false isolation event where an ESX host is incorrectly identified as being offline, triggering an isolation response.
• DRS load balancing is not instantaneous, so balancing virtual machines with rapidly oscillating load with more consistent VMware HA and DRS Capacity Planning workload characteristics will likely require a larger cluster to ensure a more rapid redistribution response.
• There are also performance considerations as to the number of hosts that should concurrently access a single LUN, as well as storage IO considerations. Too many hosts with concurrent access to a single LUN can inflict a performance penalty due to LUN-level SCSI reservations associated with virtual machine file operations and LUN metadata updates.
• Although an HA cluster can contain up to sixteen hosts, recommended cluster sizes generally fall somewhere between eight and twelve, based on the individual organization’s needs.
• VMware ESX host hardware in the cluster should be as homogeneous as possible, specifically in terms of memory capacity, CPU clock speed, and core count. Because DRS relies on VMotion to migrate running workloads, all hosts must be VMotion compatible
• It is best to have hosts of the same spec so you don’t create a skewed HA Failover situation.