Hacker News Reader

Every 100+ TB ZFS pool I have is optimized for streaming reads/writes. IOPS are rarely a binding factor at these sizes.

Isn't that kind of the point of the article? The random reads required for metadata traversal during scrubs will kill performance for sequential reads and writes as well if you don't have the IOPS to back them up, building a pool without taking this into account will make regularly scheduled scrubs a pain point.

I schedule scrubs during idle time (10pm Friday - 6am Monday), so the metadata hit doesn't affect usage, but YMMV if you don't have an idle time. I also don't generally let them get that full.

If I was building the array new, I'd probably use a special VDEV for metadata, but they didn't exist when I built my last one of these.

Disk failures don't neatly wait for idle times so investing in a special vdev (which can be added to an existing pool as the article points out) should be well worth it.

Reply on news.ycombinator.com