It appears deleting VM with checkpoint from SCVMM 2012 R2 is extremely slow.
There is the configuration:
one SCVMM 2012 R2 running on Windows Server 2012 R2
one Windows Server 2012 R2 as file share server (SMB), stores all VM files. The share is managed by SCVMM . uses local SSD disk. two gigabit ethernet ports connected to a switch.
two hyper-v hosts running Windows Server 2012 R2, non-clustered, managed by SCVMM, each hosts has three gigabit Ethernet ports connected to the switch.
Everthing works including live migration. Here is the problem:
When I deleted a VM with checkpoint, it appeared that SCVMM/Hyper-V merges something, it took a lot of time on reading each AVHDX and write to VHDX, this results two problems:
1. It took a lot of time and bandwidth. I am about to delete the VM (and all its files). Why SCVMM/Hyper-V wants to merge all files? The deletion took 29 minutes for a 36GB VM. This VM is just plain standalone VM, no usage of template.
The expected result is to just delete all VM files and remove VM, don't do merge. (BTW: VMWare doesn't do merge, the deletion is done in a few seconds))
2. although all data file are on the file share, the hyper-v host still reads each file from file share, then write to another file on the same file share. It consumes significant network bandwidth. Can they offload the work to file server?
The expected result is that Hyper-V offload the transfer to file share server. It should not create huge amount of network io.
3. The merging operation is incredibility slow. around 24MByte/sec data being copied.
The hyper-v has 2xgigabit links, benchmark shows the file share access can easily go 200Mbytes/sec, and the file server side is SSD drive with 200+MByte/sec bandwidth. The expected result is if hyper-v ever merge, the speed is least 100MByte/s and hopefully go up to 200MByte/sec. 21MB/s is just painfully slow.
Did I miss anything?