chad.parris

Using VMTurbo to manually tier across storage

Discussion created by chad.parris on May 6, 2016
Latest reply on Aug 29, 2016 by chad.parris


once you start using vmturbo, you realize that in-storage-array tiering is incredibly impractical. At least for the vmdk's.

for raw-mapped luns, sure, but the vm data is simply not going to be helped by the long-polling-interval tiering most storage arrays perform.

What you need is a storage vmotion of the vmdk's when the workload calls for it, and a habitual "packing" of the highest-performing disk in the inventory.

What is desired is in an environment where you have some flash, some spinning disk that is very good, some not so good...and some horrible....you want to provide the best experience possible for all workloads and not leave anything on the table.

Meaning, NO HOARDING of the good disk.

Pack it full, day one, and tier DOWN what won't fit on the flash.

There are several ways we have considered to do this, and we are interested if anyone has a good method that works.

We have tried putting dedicated storage clusters (talking vmware here exclusively) that are all flash attached to the same cluster that also has other grades of disk available.

What we find is that in the absence of any issues on the lower-standard disks...such as hitting a trigger of utilization, capacity or latency...nothing moves.

Should we just make the flash the lowest-cost in the environment then workload will rush there?

Has anyone implemented something that works?

again, in a perfect example, say we had 100TB of flash....300TB of 15k disks and 600TB in SATA....we would ALWAYS have the flash full, the 15k disks full, and be moving low-usage workloads down to the SATA.

And, if you see my other post, tiering the stuff off the sata onto the freezer and then purging it.

Outcomes