zfs send/receive and ARC/L2ARC

[efinley1272]

I'm building a system with two backup servers (A and B).
I'll have 10+ hosts pushing datasets to backup A via zfs send.
Backup A will then push all of those datasets to backup B using zfs send.

My question is - will backup A write everything that it receives via zfs receive into the ARC/L2ARC so that the subsequent send to backup B comes from ARC rather than disk reads?

Thanks in advance,
Elliot

 

[DrKK]

Assuming you have sufficient RAM for the task, and not much else going on, then presumably "yes". Why do you care? (Out of curiosity)

 

[efinley1272]

Assuming you have sufficient RAM for the task, and not much else going on, then presumably "yes". Why do you care? (Out of curiosity)

I care because backup A is going to be very very busy. The more reads I can serve out of cache the more I can free up the spinning disk to handle writes. Does anyone know the actual answer to this question? "presumably" isn't an answer.

 

[atreadwell]

this maybe a noob question but, Does zfs populate the arc (and subsequently l2arc if there is one) from written data? or only from read data?

 

[Johnny Fartpants]

I care because backup A is going to be very very busy. The more reads I can serve out of cache the more I can free up the spinning disk to handle writes. Does anyone know the actual answer to this question? "presumably" isn't an answer.

If you're backing up using zfs send/receive then your performance limitation won't be the spinning disks but the replication stream itself so I would say it doesn't matter.

 

[Johnny Fartpants]

this maybe a noob question but, Does zfs populate the arc (and subsequently l2arc if there is one) from written data? or only from read data?

Both I think. For example if Im benchmarking a pool I need to write something bigger than the size of the total size of RAM if I want to see disk read performance otherwise it will come straight back from RAM at super speed.

 

[rs225]

I think the answer is "try it and tell us what happens."

The ZFS caching algorithm avoids caching things it thinks are one-time uses. If you copy a 50GB file, it will probably cache very little of it. If you are seeking around in the 50GB file, it will cache what it reads.