I have a Ubuntu server running on a vm.  This is on Unraid 6.1.6.  I've allocated 4 gigs of ram (out of a total of 8 gigs) to the vm and the vm image is qcow2.  I recently converted my single disk cache  to a two disk cache pool. The two disks are 1 terabyte 7200 rpm Seagates. I know ssds would speed things up, but this is what I have for now. My former single disk cache drive was formatted as XFS.  Since converting to the pool, I've noticed a slowdown in the vm's response time especially with apps that need to access the vm's MySQL database. The delay is not terrible, but noticeable.  Is this normal?

Qcow2 virtual disk hosted on a BTRFS is identified in the KVM documentation as performing badly.  You probably want a raw image to get better performance.

Thanks, I'll try converting to raw and see what happens.

Thanks, I'll try converting to raw and see what happens.

Be interested to see if it helps.  Seen it mentioned in the KVM documentation, and in a post by Jonp but not sure of the actual difference it makes.

 

It would be nice if someone could do controlled performance tests that compare performance with different virtual disk image types (.qcow2, .img, .vdi, .vmdk) and the host file system (BTRFS or XFS). 

 

One problem with the raw (.img) format is that it tends to allocate all the space up front, while some of the other formats only allocate space as it is used inside the virtual disk.  Also it does not (as far as I know allow) for taking snapshots.  The dynamic allocation can be particularily useful when you want to set up multiple VMs hosted on the same physical disk and do not know what image size to set up.  With the formats that allocate space dynamically at least you only end up using what you actually need even if logically you have over allocated the space at the virtual level.