Re: Problems with SMR hard drives
>> AFAIK their performance for RAID is no worse than for non-RAID uses.
>> There might be specific RAID uses where they suck more (apparently RAID
>> as done by ZFS is among them), but according to the above tests it's not
>> much worse at RAID than at other things.
> Issues I have personally seen with these types of drives put into a
> conventional RAID environment:
> - pitiful random write performance once internal drive caches are exceeded
AFAIK this applies equally to non-RAID setups. It's just part of the
performance profile of SMRs: don't do random writes.
> - extremely slow rebuild/scrub speeds sometimes ten times slower than
> what I would expect from CMR drives.
That was not the experience reported in the link I sent. And I'd expect
a RAID rebuild/scrub should result in large sequential operations, which
should not trigger the usual SMR weakness. Maybe it depend on specifics
of how the rebuild/scrub proceeds?
Note also that the performance weaknesses of SMRs are somewhat similar
to SSDs, so changes to operate in larger chunks to improve SSD
performance would help SMRs as well, so maybe lots of software has
"naturally" evolved to do it in way that doesn't fall into those traps.
> Taken as a whole it's definitely not how I would design a storage system
> like most RAIDs, and it's pretty easy to not do that.
Clearly, you'd do it only if the cost-per-byte is an important criterion.
Stefan
Reply to: