VMware provisioning type for MSA2040 with virtual pool

[Biite]

Hi there,

I'm searching myself silly for best practices for provisioning VM's on VMware ESXi 6.0, my question is as follows:

MSA 2040 config: 1 virtual pool with 2 volumes (2TB), consisting of 2 400GB SSD's and 6 600GB SAS's.
VMware ESXi 6.0 via FC and SAN switches

What do I choose for provisioning on the MSA in VMware??

• Thin: seems not logical as virtual pools are thin provioned on the MSA. Thin on thin is not wise in my opinion.
• Thick Lazy Zeroed: seems logical as this is full provisioning on VMware, and thin on the MSA (with virtual pools)
• Thick Eager Zeroed: seems not logical. This is also full on VMware, but what does the MSA do with the zeroing process??

[SUBHAJIT KHANBARMAN_1 · Accepted solution]

In general I would say you should refer the below best practice guide,

https://www.hpe.com/h20195/v2/getpdf.aspx/4aa4-7060enw.pdf

Regarding Thin, Thick Lazy Zeroed and Thick Eager Zeroed, it depends on your business requirement. If you look for performance then Eager Zeroed Thick Disk almost 25 % faster in first-write performance compared to the Lazy Zeroed. Also Eager Zeroed Thick Disk almost 45 % faster in first-write performance compared to the Thin Disk.

If you concern about space reclaim then UNMAP comes into picture for Thing provisioned volume only where after deletion of data or complete VMDK file ESX need to tell MSA that you also reclaim physical space now.

In case of Thick all space already Zeroed out so from MSA perspective it's always occupied space. Now in VMFS level whatever you are doing doesn't matter for MSA.

[Biite]

Read it but the best practice guide mentions nothing about provisionng on VMware unfortunately.

Did a test in our Nordics environment, our standard is now Thick Lazy Zeroed for the reasons mentioned earlier:

• Does full provisioning on VMware, so no overprovisioning there
• Respects the thin provisioning on the MSA, so we can use the thin provisioning with virtual pools

First write Performance does not seem to be a problem at the moment.

thanks for the explananation