Will a tweak/increase of PVSCSI Queues increase performance so much that its worth it?

[SEC08]

Hello!

Going to upgrade our VMs with the controller PVSCSI but im seeing that you can tweak the queues of this controller.

I cant find what the standard queues is of mentioned controller, only finding default ones 64 and 254 from 2015 but since im using ESXI Vsphere version 7.0.3 im thinking the defaults mustve changed.

VMs are running SVN, Filestorage, Image server etc.

My questions are:

Does anyone know what the default queues is on the 7.0.3 version? As I only find KBs from 2015.

As the subject, increase these queues on device and guest OS, will it make that much of a difference in performance that it will be worth it?

Thanks a lot!

[vbondzio · Accepted solution]

If you are talking about https://kb.vmware.com/s/article/2053145, then those figures are still (mostly) accurate. The Linux device queue depth was increased from 64 to 254 by default around the 3.15 time frame, the adapter depth wasn't touched. The depth on Windows wasn't changed for either. If you are currently seeing queuing on a PVSCSI device / adapter and your storage can handle the IO, go ahead and increase it, else, don't. Note: the queue depth for devices is defined in their respective drivers and those ship asynchronously from ESXi with Tools or upstream from the OS vendors (not that it matters much in this example).

[vbondzio · Accepted solution]

Actually, there are a couple more changes that the KB article doesn't mention like the maxing of the adapter queue for HWv14 VMs by default. Let me look into updating it.

[SEC08]

Yes i am ^^ . Alright, the adapter n storage can handle it but dont know yet if its queuing as my users aint using the servers enough yet.

Thanks 🙂

[SEC08]

Aight, that would be amazing if you could. Helps a lot in upgrades.

Thanks a lot.