On Friday, May 1, 2015, Andrew Gabriel <[email protected]> wrote:
> On 01/05/2015 07:13, Nick Tan wrote: > >> Hi all, >> >> Has anyone tried using SMR disks with ZFS? I bought a Seagate 8TB SMR >> disk >> and put it in a esata enclosure for my backups. I found that zfs send >> would cause the disk to go offline. My guess is that zfs send is too fast >> and fills the drive write cache. >> >> I tried again with just rsync and this worked fine. >> > > How did you setup the drive? > What filesystem, or just writing to it serially like a tape drive? > > SMR disks have some interesting issues with recording, particularly when > writing non-serially as most filesystems normally do. Since there's no SMR > support in Illumos, I presume you ran the drive in Drive Managed mode - > this makes it look like a standard random access drive. However, like a > flash drive, it will actually be laying the data on the drive out very > differently from what the host OS/filesystem imagines. Also like a flash > drive, it will have to do some housekeeping and move blocks of data around > on the disk and/or re-record large ranges of data previously written, so > performance from the host system may appear very mixed, including some i/o > requests which take long enough that with a standard magnetic drive you > would assume the drive is dying (probably why you saw the drive reported as > going offline). This may be fine for archival/backup data (providing the > host system knows to allow a long time for i/o), but is less likely to be > good for normal filesystem use by applications. > > There are better ways of driving SMR drives, but they require support in > the operating system and/or application. One of the uses they are more > suitable for is a key/value object store, because drives can implement the > object store layer entirely in the drive firmware hiding the real layout > from the system, and data is accessed entirely by using keys. > > -- > Andrew > > _______________________________________________ > openindiana-discuss mailing list > [email protected] > http://openindiana.org/mailman/listinfo/openindiana-discuss > I set it up as a single drive zpool. As it's just being used for backup I figured it would be ok since the use case is large sequential writes. I think though that zfs send just overwhelmes it. I disconnect it and take it offsite so I'm not too worried about other I/O to it. I'll reconnect it once a month for a new sync. However since zfs is CoW it should be ok on subsequent rsyncs. This system is replacing my lto-2 tape library and so far it's been a good experience. It is certainly faster than tape! _______________________________________________ openindiana-discuss mailing list [email protected] http://openindiana.org/mailman/listinfo/openindiana-discuss
