Different IO Performance in template and in linked clone replica

[sappomannoz]

Hello,

I'm using locally attached SSD in my view POC. I have investigated the Read I/O performance of a linked clone VM and it seems to be much slower (random I/O) as the base image, both residing on the same SSD

Here are the results for the base image

and here the results of the linked clone

Is this expected or do I have some issues with my setup?

BTW the speed of the disposable disk is ok:

[Najtsob]

This is normal.

When accessing C:\ disk of the linked clone, hypervisor must read and write to both replica AND delta disk. So using rule of thumb 100 IOPS inside guest OS translates to 200 IOPS on the physical disk.

[sappomannoz]

Hello Najtsob,

thank you for your reply, but is this also valid for a read test like in my case? Shouldn't it read directly from the linked clone replica, since afaik HD Tune used pre-existing data for the read io test.

Kind regards

Cristiano

[iannobleuk]

Is the base image hardware version 10? Sesparse is a new disk format that gives better performance for linked clones, which is the format used if you create the linked clones off of a base with a virtual hardware version that supports it (and are using ESXi 5.1 or higher).

Change the hw version, recreate the linked clone(s) then redo the tests

Ref:

Re: Why Linked Clone disk so slow?

SEsparse in VMware vSphere 5.5