Page 1 of 1

log_spacemap in practice

PostPosted: Thu Jan 13, 2022 8:50 am
by RJVB
Hi,

Can anyone give feedback on using the log_spacemap feature in practice, both positive and negative. Experience from the Linux domain more than welcome too - in fact I run a Linux box from a ZFS pool which evidently sees considerable fragmentation so every bit of *safe* performance gain is a plus.

Thanks!

Re: log_spacemap in practice

PostPosted: Mon Jan 17, 2022 12:49 am
by lundman
Not really look at it myself, but

Christian Schwarz 3 minutes ago
I had a very positive experience when I upgraded to FreeBSD 13 on a 4-year-old 2HDD mirrored pool. I observed a 4x reduction in IOPS per txg sync.

Re: log_spacemap in practice

PostPosted: Mon Jan 17, 2022 12:43 pm
by RJVB
Christian Schwarz wrote:a 4x reduction in IOPS per txg sync.


I saw a study that presented gains in the same or similar units. How do those translate to real-world observables?

Is there a better place to ask this, where I stand more chance of getting feedback? I tried on IRC #zfs, but no one ever seems to be there (when I am), and I (ahem) couldn't figure out how to use the other channel (?) on "slack".