Context : We want to run MySQL databases on Proxmox (kvm).
Problem : MySQL is quite slow on kvm. Compared to native speeds or Xen speeds, it is up to 5x the latency.
Setup : 3-node Proxmox cluster. On 1 node, we add a dedicated SSD disk for Mysql, and use LVM to mount it in the VM.
VM disk info :
scsi3: vm-data-lvm:vm-104-disk-2,iothread=1,size=150G
scsihw: virtio-scsi-single
cache: no-cache
Thick provisioning.
Tool used to test :
ioping -S64M -L -s4k -W -c 10 .
Latency Results in VM on XFS : (ext4 is about 200us higher average)
min/avg/max/mdev = 562.4 us / 663.2 us / 742.1 us / 51.1 us
Same LV, mounted on the host :
min/avg/max/mdev = 120.5 us / 149.5 us / 183.3 us / 20.9 us
Tests in a Xen VM is about as fast as native speeds.
I was expecting something like 5% to 10% slower/higher latency, but this is a bit extreme
For info : We tested a LOT of settings. XFS/EXT4, cache settings, but could not get close to native speeds.
We also tested with sysbench using oltp_write_only and the results were similar.
Is this something we should expect from kvm ?
Anyone seeing better results ?
I guess not knowing if this is normal or not is a big issue here. Any help would be greatly appreciated !
Problem : MySQL is quite slow on kvm. Compared to native speeds or Xen speeds, it is up to 5x the latency.
Setup : 3-node Proxmox cluster. On 1 node, we add a dedicated SSD disk for Mysql, and use LVM to mount it in the VM.
VM disk info :
scsi3: vm-data-lvm:vm-104-disk-2,iothread=1,size=150G
scsihw: virtio-scsi-single
cache: no-cache
Thick provisioning.
Tool used to test :
ioping -S64M -L -s4k -W -c 10 .
Latency Results in VM on XFS : (ext4 is about 200us higher average)
min/avg/max/mdev = 562.4 us / 663.2 us / 742.1 us / 51.1 us
Same LV, mounted on the host :
min/avg/max/mdev = 120.5 us / 149.5 us / 183.3 us / 20.9 us
Tests in a Xen VM is about as fast as native speeds.
I was expecting something like 5% to 10% slower/higher latency, but this is a bit extreme
For info : We tested a LOT of settings. XFS/EXT4, cache settings, but could not get close to native speeds.
We also tested with sysbench using oltp_write_only and the results were similar.
Is this something we should expect from kvm ?
Anyone seeing better results ?
I guess not knowing if this is normal or not is a big issue here. Any help would be greatly appreciated !