Ran in to an issue before Christmas whereby ESXi would take over 10 minutes to boot. The issue appeared to be with regards to certain LUNs timing out during the rescan operation.
Upon further investigation, the LUNs causing the issue were RDMs connected to a MSCS cluster. The workaround for this was to set the Scsi.UWConflictRetries from 1000 (default) to 80. Whilst this shortens the boot time, it’s still relatively long and really worrying when utilising DPM.
KB1016106 covers this topic in full.
1000 times by default!
It is 80 by default for an ESX3.5 hosts with a recommemded value of 10 (seen on VMware forums) when the host takes a long time to boot up.