Our View environment consists of 200 linked clones made from a master Win7, 32bit machine. The master has a single 24GB disk that was using the LSI Logic SAS SCSI controller. To improve performance and cut down on CPU load we decided to change over to the Paravirtualized SCSI controller. After making this change on the master, we recomposed the pool. Once the recompose was finished I expected to look at the settings of any of my linked clone machines and see them also using the Paravirualized SCSI controller. Much to my surprise I got mix and match as shown below. When linked clones are made it also creates a small 20MB disk along with the 24GB disk. I got two Paravirtual controllers, on Paravirtual and one LSI, and two LSI controllers. Any idea why?
I kind of chuckled at that mix mash of results Are they all from the same pool? What kind of results do you get if you simply remove a VM and let it get recreated vs a recompose?
They are from two different pools
My student pool gets either:
My adult pool gets either:
I just removed a VM that composed with two LSI adapters and it came back with two Paravirtual and it was in the Student pool. Not sure how the compose operation ends up with such mix and match. We are running ESXi 5.1 and View 5.1. We are soon to upgrade to View 5.2, maybe that will fix the problem.
Any idea if there is much of a performance boost in the View environment when running the Paravirtual adapter over the LSI?
Did you ever get an answer on this. I am doing that as well as it was recommended to us to stop some LSI_SAS errors.
I now have two SCSI Paravirtual when the master has only one.
This second contoller is by design. Horizon attaches its own disks (e.g. internal (~20mb), disposable, persistent) to a new controller that is created during compose operation.
The second controller should match the type of the first, this is what was the original issue.
I think this could be related to recompose and not deleting the old vms that had an other controller type earlier. Not all virtual hardware components are renewed during recompose actions.
Best regards
Andreas
I did realize that too. And it is consistent in creating two PVSCSI for me and working well. I thought it was contributing to my problem with windows creating a temporary paging file. But that was a separate issue.
I have a Horizon 7 environment where the parent VMs were initially built with only PVSCSI controllers. I can confirm that two PVSCSI controllers is the expected behavior.
Does anyone actually see vm improvements or just overall cpu reductions. I've been following vmware's best practices pretty closely and I've been pretty happy, but they still have the lsi sas in the documentation I've been reading.
This touches a bit on SATA vs LSI SAS vs PVSCSI -
Performance Issues Due To Virtual SCSI Device Queue Depths | Long White Virtual Clouds