cancel
Showing results for 
Search instead for 
Did you mean: 
sabinakhanal
Mission Specialist
Mission Specialist
  • 492 Views

VDO testing

Hi team, I am currently trying to do some testing with VDO. I created a volume group, a logical volume of type VDO using the following commands:
vgcreate vdovg /dev/nvme0n2

lvcreate --type vdo --name vdo0 --size 1T --virtualsize 3T --compression y --deduplication n --config 'allocation { vdo_write_policy="sync" }' vdovg 
(I just need compression for this testing purpose)

now I am trying to run fio against the logical volume:
sudo fio --filename=/dev/mapper/vdovg-vdo0 --rw=write --direct=1 --bs=8K --ioengine=libaio --runtime=300 --numjobs=1 --time_based --group_reporting --name=seqwrite --iodepth=1

I am getting like 27.2K IOPS and 36 microseconds latency. How is this possible for QD1? 
when I run a test against the block device nvme0n2, I get around 6K IOPS and 160 microseconds. How can there be such a huge difference?

3 Replies
Trevor
Starfighter Starfighter
Starfighter
  • 450 Views

Let me begin by gathering one piece of information for starters.  When you run
fio on both devices, are you running an apples-for-apples test (e.g. using the same
block size - in your example 8K)?

I'll have more questions down the road, but I'd like to start here.

Trevor "Red Hat Evangelist" Chandler
sabinakhanal
Mission Specialist
Mission Specialist
  • 449 Views

Hi Trevor, thank you so much for your response. For this scenario, I just have one VM where I am creating this logical volume on an nvme disk. I am logged in to the VM and am running fio against the logical volume on the disk. 
and yes, I was running the tests on the logical volume and the disk with the same block size and same other parameters

sabinakhanal
Mission Specialist
Mission Specialist
  • 399 Views

@Trevor , also adding few more details. 
Chapter 4. Testing VDO performance | Red Hat Product Documentation
I am following the fio tests in this documentation and these are the results (fio against the logical volume created using vdo):
QD1 seq read: 5448 IOPS, 187.11 us | randread: 3740 IOPS, 276 us
QD1 seq write: 28.5K IOPS, 36 us | randwrite: 28K IOPS, 36 us

I think I am just trying to understand why is there such a huge difference between read and write numbers. I have never seen 28K IOPS and such low latency for qd 1 before so trying to understand if I am doing something wrong. 
Please let me know if you have any other questions. 

Join the discussion
You must log in to join this conversation.