Hi Experts,
What is the best practice when comes to datastore.
like whether we should have a less numbers of big LUNs or more numbers of small LUNs.
How many Vms can be placed in a LUN? (not vmware maximum, need real time best practice)
Is combining multiple LUN in a datastore is best practice?
Thank you.
Is combining multiple LUN in a datastore is best practice?
I would suggest to have single LUN for each datastore..
Check the below for similiar discussion
http://communities.vmware.com/docs/DOC-10990
Best practices for creating datastores
VMBuddy201110141 wrote:
Hi Experts,
What is the best practice when comes to datastore.
like whether we should have a less numbers of big LUNs or more numbers of small LUNs.
Something in the middle. 1 giant data store is a bad idea for performance reasons. 100 data stores is a bad idea for management. Choose a number > 4 that meets your needs in terms of business requirements.
How many Vms can be placed in a LUN? (not vmware maximum, need real time best practice)
There's no real limit here. As long as your array supports VAAI, put as many VMs as you feel comfortable with and that the array can support from an IO perspective. Some arrays with clever tiering and wide striping can do hundreds. Some cant. Best to ask your specific vendor.
Is combining multiple LUN in a datastore is best practice?
Generally, no.
Thank you.
After doing tonns of storage migration for clients, here is my take:
Design your storage understanding the Virtuals. Understand the application workloads and Env.
If possible avoid extents in a Datastore, Keep a one-to-one mapping. One Lun - One Datastore.
Doing Thin Provisioning has its own downsides (its more fragemented) so make sure your VMs does not run any specific loads which rely a lot on data. Ofcourse, It lets you put many VMs per DS.
Even If you keep it Thick, try to use 80/20 rule for keeping safe.
One of the reasons I think of keeping a lesser no. of VMs on a datastore would be type of storage, its setup, no. of paths etc. so that in case you dont have features like sDRS, you keep your VM/DS ratio balanced and dont sacrifice a lot of VMs if all path failure occurs.
Hope this helps.
Thanks,
Aman