Poor LVM Performance

M@ario

Member
Dec 23, 2010
34
0
6
Hi!

I installed 1.7 on two nodes and updated to Kernel 2.6.35 (as I plan to use KVM only).
My storage is a large 1.8TB FC Array. I created a PV and VG (VM_Data) and added the storage as LVM group in Proxmox. Problem is that I got very poor I/O performace in the KVM VMs.
So I added a test LV and tried some dd on the LV.

dd if=/dev/sdc2 of=/dev/VM_Data/test bs=102400 count=10000
1024000000 Bytes (1,0 GB) kopiert, 72,8614 s, 14,1 MB/s

When i dd to the FC SCSI drive directly I get
1024000000 Bytes (1,0 GB) kopiert, 0,486052 s, 2,1 GB/s

Can anyone explain this to me?
 
Hi!

I installed 1.7 on two nodes and updated to Kernel 2.6.35 (as I plan to use KVM only).
My storage is a large 1.8TB FC Array. I created a PV and VG (VM_Data) and added the storage as LVM group in Proxmox. Problem is that I got very poor I/O performace in the KVM VMs.
So I added a test LV and tried some dd on the LV.

dd if=/dev/sdc2 of=/dev/VM_Data/test bs=102400 count=10000
1024000000 Bytes (1,0 GB) kopiert, 72,8614 s, 14,1 MB/s

When i dd to the FC SCSI drive directly I get
1024000000 Bytes (1,0 GB) kopiert, 0,486052 s, 2,1 GB/s

Can anyone explain this to me?
Hi,
you meassure caching!

To get "real" data use something like this:
Code:
# delete caches
echo 3 > /proc/sys/vm/drop_caches

dd if=/dev/zero of=/dev/VM_Data/test bs=1024k count=8192 conv=fdatasync
To get the read-performance you can also create a filesystem on the -test-lv, mount this to /mnt and do a "pveperf /mnt".

Udo