Hi,
I'm testing out PBS. I have bought a single EXOS 20TB HDD as my local backup repository. That repository should also backup to tape. I've got it all working but I noticed slow LTO speeds (~85MB max while it should get to 120MB/s(*)). While investigating I noticed the CPU doing a LOT of wait states while writing to tape. Those wait states come from the HDD 100% READ busy. It's constantly reading and only spitting out ~85MB/s so definitively the bottleneck. I have tried some minor ZFS tweaks which helped a little and also reformatting the repository as ext4 which also improved speeds, but still 85MB/s tops and still a lot of I/O wait states in the CPU. My best guess is that it's due to the chunk size and it can't handle the IO. The backup server is otherwise doing nothing else.
The backup server is a physically separate server. The OS runs from an NVME SSD, the repository should go on the 20TB single SATA drive.
I've got about 20LXC containers of moderate size (1-10GB max) and one ~4TB NFS-server with JPEGS and RAW images.
So my question: can I change the chunk size on the backup repository to something larger and see what it gives? If so how?
I reckon buying an SSD will also fix it but my backup repository is too large to put it on SSD's (home lab test, no budget )
(*)LTO-4 I know, not officially supported but everything seems to work well, backups work, ejecting works, reading LTO RFID stuff works so I guess not a firmware problem.
I'm testing out PBS. I have bought a single EXOS 20TB HDD as my local backup repository. That repository should also backup to tape. I've got it all working but I noticed slow LTO speeds (~85MB max while it should get to 120MB/s(*)). While investigating I noticed the CPU doing a LOT of wait states while writing to tape. Those wait states come from the HDD 100% READ busy. It's constantly reading and only spitting out ~85MB/s so definitively the bottleneck. I have tried some minor ZFS tweaks which helped a little and also reformatting the repository as ext4 which also improved speeds, but still 85MB/s tops and still a lot of I/O wait states in the CPU. My best guess is that it's due to the chunk size and it can't handle the IO. The backup server is otherwise doing nothing else.
The backup server is a physically separate server. The OS runs from an NVME SSD, the repository should go on the 20TB single SATA drive.
I've got about 20LXC containers of moderate size (1-10GB max) and one ~4TB NFS-server with JPEGS and RAW images.
So my question: can I change the chunk size on the backup repository to something larger and see what it gives? If so how?
I reckon buying an SSD will also fix it but my backup repository is too large to put it on SSD's (home lab test, no budget )
(*)LTO-4 I know, not officially supported but everything seems to work well, backups work, ejecting works, reading LTO RFID stuff works so I guess not a firmware problem.
Last edited: