I'm currently testing out the Spaceinvader One Mastering ZFS: Easily Auto Snapshot & Replicate a ZFS Dataset and I'm seeing extremely slow transfer speed, around 10-15MB/sec. I believe it might have to do with the fact that the pool I'm replicating to has compression enabled. When I start a replication I can see the CPU usage of one single core get pegged to 100%. The core that is being used will jump to a new one after a bit, but it will always only use a single core. Is this expected, or is there a way to change something in zfs settings to utilize all cores during the compression process?

I've been following many of the incredible guides from @SpaceInvaderOne, and recently decided to set up zfs auto snapshots and replication from my pool to a zfs formatted drive on my array using his script. Everything seems to be working correctly, but the speed seems extremely slow, averaging around 10MB/s. It took over two hours to replicate a dataset around 83GB. Does this seem right? I notice that while the replication is happening, one of my cores on my processor is pegged at 100% the entire time and core that's pegged at 100% will switch at random.

 

5 hours ago, nicduv said:

around 10-15MB/sec.

Is that for the array or a pool?

8 hours ago, JorgeB said:

Is that for the array or a pool?

This is replicating from my raidz 4 nvme drive cache pool to a single disk HDD zpool in the array.

 

The cache peaks at 25MB/s read sometimes while running the replication. The array drive sits around 10MB/s write.

 

I don't think it's actually anything to do with compression, as I disabled compression on the array drive and am still seeing the same slow speed and single core of the CPU getting pegged.

There's currently a known issue with the zfs write speed when used in the array, pools are not affected

Got it. I'm still on 6.12.14, is this issue still present in version 7?

Yep.