Improve TimeMachine performance?

[matram]

I am running three different backups on my MBP (yes, I like to play it safe):

• Monthly CCC to HDD
• TM to USB-C connected HDD - which gives me about 80 MB/sec average backup speed
• TM to network connected Samba server on Intel NUC with SSD drive - gives 6 - 8 MB/sec
• (used to run) TM to AirportExtreme with USB drive - which gave 3 - 4 MB/sec

What confuses me is the slow performance of TM to a network share. To compare:

• File copy to the samba share using Finder gives 14 MB/sec, i.e. double TM performance
• Secure copy (scp in terminal) from the MBP to the samba share gives about 70 MB/sec

The network connection is 1 Gbit cabled via a Cisco ASA-5506. The samba server has been upgraded to latest Debian 10 and TimeMachine is now run over SMB protocol instead of AFP. Neither disk, network or cpu shows any real load when running TM (a few percent).

Anyone has any good suggestions for how to speed things up?

 

[cmstuber]

https://******.com/?q=time+machine+speed
***** = lmgtf_y

Seems like it's intentional by apple to keep things slow so it feels more like a background low priority task rather than you feeling the cpu and network hit every single hour. there also appears to be a way to change that with some commands.

 

[matram]

I am aware of the possibility to increase the priority of the backupd process. I have tried that, and it has no effect. Backupd fully loads one thread when it is preparing backup, i.e. figuring out what files it need to copy. Once it starts copying, backupd only uses a few percent CPU as the copying is done by kernel_task, that runs at about 15% during copy.

As mentioned when TM runs agains a locally connected disk the speed increases by more than 10x, so it is not that it cannot run fast.

 

[kohlson]

If I understand your question correctly (why is a TM backup slower than a local/network disk copy), I think TM is doing several more things than "copy." For example, moves, adds changes, and deletes; possibly encryption.decryption - then determine the changes and copying accordingly. And as mentioned above, it tries to do this with minimal impact.

Perhaps a closer comparison would be something like Carbon Copy Cloner and/or SuperDuper. They both spend some amount of time determine what needs copying before actually copying it.