Volume block size and LUN snapshot performance

IF your questions pertains to using the file server performance, transfer issues, jumbo frames tuning, post it here.
Forum rules
We've moved! Head over to Synology Community (community.synology.com) to meet up with our team and other Synology enthusiasts!
I'm New!
I'm New!
Posts: 1
Joined: Fri Sep 29, 2017 9:05 am

Volume block size and LUN snapshot performance

Unread post by jcguillain » Fri Sep 29, 2017 10:20 am

The purpose of this post is to share our experience with LUN snapshot performance.

We use LUNs on a RAID6 volume. These LUNs are shared using iSCSI to Linux virtual machines (Xen) system and data volumes formated in ext3/4.
We intensively use replication between two RS3614 appliances, and we have serious performance issues during the snapshots done before each replication.

One of our idea to improve these performances was to change the block size of the volume. (We originally used 8k block size.)

We measured the time needed to take a snapshot of a 512G LUN (393G used) with different volume configurations :
- 64k block with SSD cache
- 8k with SSD cache (our original configuration)
- 64k block
- 16k block
- 8k block

And with different read/write load.
- deletion of an old snapshot before the test, and R/W load
- R/W load
- deletion of an old snapshot before the test, no R/W
- no activity

The filesystem used on the LUN is ext4 with 4k block.
The read/write load is generated by the parallel read and write of random files of 24M.
(We measured the time when an old snapshot was first deleted, because it slows down the creation of a new one during several minutes)

This is the graph :


There is no doubt that the 64k block size improves significantly the performances, and we are currently migrating our volumes using this size.


Return to “File Server Performance”