vSphere Cluster Design
ESXi hosts and their resources are typically pooled together into clusters. These clusters contain the CPU, memory, network, and storage resources available for allocation to virtual machines. Clusters can scale up to a maximum of 64 nodes in vSphere 6.0 and can support thousands of virtual machines.
When designing clusters, there are two basic strategies:
• “Scale-out clusters,” which have fewer ESXi hosts, but the architecture will end up with a larger overall number of clusters configured.
• “Scale-up clusters,” which each have more ESXi hosts, but a fewer number of overall clusters.
Therefore, the design decision is whether to have many smaller clusters or fewer larger clusters.
As with “scale out” or “scale up” at the host level, each approach to cluster design has a number of advantages and drawbacks, but the decision will be driven by a number of potential factors as described in the following table.
Table 13. Advantages and Drawbacks of vSphere Cluster Design Options
Advantages | Drawbacks |
Scale-out clusters (fewer numbers of hosts per cluster) |
• You can be less concerned about staying within the vSphere HA cluster virtual machines per host maximum. | • Depending on the admission control policy, more resources might be reserved for failover, reducing the amount of resources available to host virtual machines. • Reduced VMware vSphere Distributed Resource Scheduler™ (DRS) migration options. |
Scale-up clusters (larger numbers of hosts per cluster) |
• Fewer resources might be reserved for failover, increasing the amount of resources available for virtual machines. • If DRS is enabled, more hosts offer greater migration choices and more opportunities to achieve a better workload balance across the cluster. | • There is DRS algorithm overhead on vCenter Server due to the larger number of calculations required. • Depending on the storage employed, presenting LUNs to large numbers of hosts could affect performance when locks are placed on datastores. |
Other general factors, including the following, might also play a part in the design decision:
• Licensing (consider Oracle)
• Shared storage presentation or SAN zoning
• Redundancy (N+1, N+2, N+3 and N+4 models)
• CPU architectures (Intel/AMD)
• Network connectivity
• Total numbers of virtual machines