We are having continual speed issues with a repository on an XFS filesystem.
This repo is set up on a linux machine. The filesystem is an iSCSI mount from a Synology box running DSM 7.2.2-72806 Update 3. I know, Synology bad. Synology is not the issue here.
The repo is used as an off-site backup location for 7 clients.
The issue is write speed to this repo.
I tested speed from a device on the same network using CrytsalDiskMark.
Sequential write speed was as high as 17 MB/s.
Random writes were as slow as 0.36 MB/s.
These speeds are consistent with what I see in Veeam. Brutally slow.
We recently had to run an active full on a server. It took 277 hours to transfer 1 TB over a 1 Gb/s link.
I did a second test. I configured a new iSCSI target and created a new LUN on the same RAID as the repository. I mounted this NTFS on a Windows box on the same network.
Sequential writes were as high as 86 MB/s.
Random writes were much better at 4 MB/s.
In general, the speeds were 7-10 times as fast with NTFS on Windows as compared to XFS on Linux.
When this repo was first set up in January, the speeds were much faster.
I checked the fragmentation on the filesystem and I found fragmentation over 99% and >5000 extents per file.
xfs_db> frag
actual 6447827, ideal 1150, fragmentation factor 99.98%
Note, this number is largely meaningless.
Files on this filesystem average 5606.81 extents per file
Currently the filesystem is 32 TB with 76% used.
I attempted to defrag the files. After 4 continuous days, the defragmentation had completed about 10% of the files. I suspect that this will never complete.
So my questions are:
- What is the Veeam recommendation when dealing with fragmentation on XFS?
- What are people doing in the real world to deal with it?