Performance when pooling LUNs
Posted on 2011-02-15
We currently have a small ESX cluster running off an iSCSI san with 2 nodes.
We are looking at an expansion this year of 2 more nodes.
I plan on making a number of smaller LUNs across all the nodes to take advantage of the extra spindles, but my question is does anyone have an experience with pooling or using metaluns in VMware? I see I have the option of pooling multiple smaller LUNs into a larger one using the storage console. I can see how that would be easier from a management perspective but I have a few questions with how that works.
Do VMDK files get spread across the multiple smaller luns that make up the pool? do the snapshots? or do the files stay together?
What happens if we have a node failure? do just the VMDK files in the small LUN go offline or do we lose the whole pool?
How does VMware allocate space in the pooled lun? Does it fill one of the smaller luns first and then move onto another or does it attempt spread the VMDKs out?