Any experiences with large record sizes? (8M+)

[probain]

We got support to easily utilize large record sizes, up to 16M.
I'm interested in if there are any experiences that anyone would like to share, after having used and experimented with them. There aren't many such stories online. So I'm hoping that people care to share.

Thx

 

[Ericloewe]

Good question. I stick to 1M out of a non-specific feeling that 2M+ is rarely used, even if long supported at this point, but 16M would be more betterer.

 

[probain]

After doing a couple of simple tests. Copying over ca 9TB of video. I've noticed a couple of simple observations.
There was a small efficiency gain in compression, going from 1M -> 16M. We're talking single digits percentage 1-5% here.
Rsync transfer speeds plummeted. From roughly 200+MB/s on 1M destination, it would drop down to 50-80MB/s when using 16M.

System hasn't gone through scrub yet. So I don't know if it has had any impact on scrub duration either.

However. After these initial "observations". I'm going back to using 1M for the video dataset. Feels like a reasonable balance between record size and speed.

 

[RetroG]

I've been using 16M recordsize (since creation) on my second (much smaller) TrueNAS Scale box, it currently has a 64T raw 8x 8T wide raidz2 pool. the data is WORM, and nearly all of it is video content, with some SMB shares of random files.

honestly it's been rock-solid, no real performance/resource deficiencies that I've noticed. though also as mentioned above, there is little reason to change and re-write existing data this way (as the gains are simply too small to justify).