What is "delete" in disk latency reporting graph?

[charvey]

I have a RAIDZ2 pool. When running a read test (dd, not cached) I get about 110MB/s on 5400rpm consumer SATA3 drives.

I looked at the disk reporting page (busy and latency) during the test and found the following:
da0, da2, da3:
Busy 40%, Read latency: 1-2ms, no write or delete

da1:
Busy 100%, Read latency: 6-7ms, delete: 10-40ms, no write.

Under almost no load the delete latency for all drives is 0, except da1 which hovers at around 1-2ms.

What is a "delete" operation? da1 is a new WD20EZAZ with 1560 power on hours and no errors on a long self test. I have much older drives that I would expect to be slower than this one. I don't keep new or expensive drives in my pool.

Thanks.

 

[Johnnie Black]

WD20EZAZ

SMR drive, they should be avoided with FreeNAS/ZFS.

 

[charvey]

SMR drive, they should be avoided with FreeNAS/ZFS.

Isn't SMR re-write an implementation detail and something that wouldn't get exposed to FreeNAS? It's also only something that happens on writes. Why avoid it in zfs (performance aside)?

 

[Johnnie Black]

Why avoid it in zfs (performance aside)?

They should be avoided for performance reasons, both for write speed and resilvers which are also usually much slower, if performance isn't a concern they can be used like any other disk.