VMware Cloud Community
admin
Immortal
Immortal

Performance, Management and Recoverability of Datastores

Hi All,

We are in the process of restructuring our datastores to take advantage of SRM. We have 60 vm's that have iscsi attached volumes ranging from 20gb to over 3tb. I am stuck on deciding whether to use pass-through on each of these volumes, or configure them as vmfs volumes. Secondly, if I went with moving the data to vmfs, do I create a single vmdk for each volume or do I put multiple vmdk's under a couple different vmfs volumes of around 2tb of size? I do have some concerns around the performance of multiple disks under a single datastore. However the issue I am faced with is the amount of connections on these volumes because we use Equallogic. We are limited in the amount of sessions we can establish. We also use the Equallogic MPIO which increases the connection count substantially for each vmfs volume since they are accessible to all hosts in the cluster.

Curious if others have some opinions on a strategy that would work best for this kind of situation.


Thanks

Reply
0 Kudos
1 Reply
fatbobsufc
Contributor
Contributor

From my experience it would depend on lots of factors such as...

Consider the performance on the underlying storage - more LUNs could potentially mean that you could balance out the IO demands on the SAN more evenly.  However, there are overheads to consider with high numbers of LUNs, hard limits and depending on the array, the increase in overheads could rule out any performance benefits.

For your 3TB disks you will need to use RDMs to get around the 2TB vmfs file size limit.

Will you be using SAN snapshotting/replication? If so go for RDMs.

Will you be using physical to virtual clusters?  If so you will need RDMs.

My personal choice is that I always prefer to use vmdks if I can rather than RDMs as I prefer the flexibility.  I always try and keep the datastores as small as possible, just making them bigger if there's a real need, preferring multiple datastores to one big one.  In many of the environments I work on, most vmdks are less the 100GB so I go with a rule of thumb of making my datastores 700GB in size.  If I need a bigger one I just create it bigger.

Not sure if this helps as the scope of your question seems quite wide...

Reply
0 Kudos