[one-users] VMFS -- VM IO Performance Problem

[chenxiang at aquala-tech.com]

Hi all,

I have a working 3.8.3 setup, with 1 front end running Ubuntu 12.04, which
also exposes two LUN's via iSCSI TGT for the ESXi node to be used as
datastores. I was able to setup VMFS datastore in OpenNebula, register
images, and instantiate VM's.

Then I did a IO performance from within the VM, with two test cases:

(1) With vSphere Client, I created a Ubuntu 12.04 VM with 15 GB disk,
install iozone inside the VM, and tested for write throughput. The VM was
installed onto the iSCSI datastore, and the write throughput I got  was
about 75 MBps.

(2) Copied the above-mentioned VMDK to the front end, and registered it as
an images in OpenNebula. Created VM template with exactly the same CPU,
memory, network, and the above-mentioned image, and instantiated the VM.
Again, ran iozone to test for write throughput, and the result was only
about 20 MBps.

The question was, both VM had identical disk images, and were running on
the same storage. Why the one instantiated by OpenNebula exhibited bad IO
performance?

Below is the VM template in OpenNebula:

CPU="1"
DISK=[
  IMAGE="VMW-Ubuntu-12.04-Perf",
  IMAGE_UNAME="oneadmin" ]
GRAPHICS=[
  TYPE="vnc" ]
INPUT=[
  BUS="usb",
  TYPE="mouse" ]
MEMORY="1024"
NAME="VMW-Ubuntu-12.04-Perf"
NIC=[
  NETWORK="VMW-LAN",
  NETWORK_UNAME="oneadmin" ]
OS=[
  ARCH="x86_64",
  BOOT="hd" ]
RAW=[
  TYPE="kvm" ]
RAW=[
  DATA="<devices><controller type='scsi' index='0'
model='lsilogic'/></devices>",
  TYPE="vmware" ]
TEMPLATE_ID="18"
VCPU="1"


Thanks a lot.

Chen Xiang