Linux Repo - Slow Transfer

[stew930]

When full backups are executed the max Processing Rate is 26%. Full backups of 20+TB takes nearly two weeks while incremental of 17GB is 12 hours. I sent logs to Veeam who said I need to look at my configuration of the Linux Repository as it appears the small file transfer rate is extremely slow. Anything I should be looking at in particular?

Currently a full backup is running and the CPU usage is very low, 4GB of 128GB of RAM used, Swap File is 8G and not used. Drive on the Linux box is 196TB XFS reflink with CRC enabled. The data is coming from another on prem server.

Ubuntu 20.04.4


Case # 05507448

[HannesK]

Hello,

max Processing Rate is 26%.

I assume that means 26 MByte/s? Yes, that's lower than my old home lab. Something is wrong.

it appears the small file transfer rate is extremely slow

which product / functionality are you using? The case sounds like Veeam Agent for Linux with file-level backup mode (the post was in the general Backup & Replication forum, which is about VMware & Hyper-V).

Without knowing the details: please use use block-based backup (entire machine / volume-based backup). That's what the agents are built for.

Best regards,
Hannes

[stew930]

Yes the Processing Rate is 26MB. For the backup to be immutable I am using a volume based backup.

[HannesK]

hmm okay... just for information: immutability also works with Veeam Agent for Linux and file-based backup. If volume based backup is used, then files are irrelevant. Volume-based backup is block-based.

I can only recommend to text every component independently from Veeam
1) test repository write speed
2) test network
3) test source read speed

As far as I see, support already did that and figured out that the disk performance is low.

[stew930]

Which is why I started the thread asking if there is anything I can check with the drive configuration assuming they are configured properly for a Veeam Linux Repository.

[bytewiseits]

We built a Ubuntu 20.04 repository and were getting abysmal speeds using Veeam (50MB/s), yet file copy/transfer/testing on the box itself would be fine (over 1GB/s). After trying many different configs and setting combinations we finally stumbled across some posts about buffered/unbuffered disk access and tried setting 'UseUnbufferedAccess' DWORD registry entry to '0' on the VBR server.

Since setting that the write speeds jumped to around 1050MB/s (proxy is now the bottleneck, not target) and is still working well. Oddly enough we also noticed our Windows 2022 REFS repos have also slightly increased in speed (not hugely but it is noticeable).

If you are getting good disk speeds when testing on the repo itself but it is just Veeam writing slow, it could be worth a test. Worst case if it doesn't make any difference you can just remove the registry entry and restart VBR/Repo to return back to the defaults etc.

[stew930]

Thank you very much bytewiseits. I added that registry entry which changed the transfer from 26MB to 500-700MB.