Reading a knowledge base post on the equallogics site I gleaned the following variables in relation to ESX that will have a factor on deciding VMFS volume size;

  1. ESX has a limit of 64 targets (volumes) per host
    • This is important in a HA / DRS environment where each host in the cluster must have access to all volumes even if the guest is currently running on another host.
    • Hence keeping volume counts down will improve scalability of the cluster
  2. ESX has a maximum queue length per target of 32 IO’s -
    • Thus more volumes means less pausing of IO opperations
  3. Certain Opperations (Start/Stop VM / Snapshot / vMotion) Require exclusive access to volume
    • Thus for a short period of time other vm’s IO is paused -
    • the more VM’s per volume the more this will impact performance
  4. VI 3.5 didn’t support MPIO so more volumes increased throughput
    • Does this matter anymore with MPIO in ESX 4? and multiple sessions per host taking advantage of all paths to the target.
  5. Don’t forget to account for snapshots in your volume size
    • vmware snapshots grow very quickly as they are an on-write copy delta of both the vm memory and it’s virtual hard disk.