Hello,
thanks for the answer.
I also searched a bit and found for this pdf:
https://oss.oracle.com/~mkp/docs/osd2008-data-integrity.pdf
It seems like all newly SAS drives you can buy not only supports the standard 512 bytes, but also 520/528 bytes. These extra bytes are checksum bytes.
(For example if you look at the datasheets from current SAS drives from HGST or Seagate you find this:
4Kn: 4096, 4112, 4160, 4224
512e: 512, 520, 528)
(For example if you look at the datasheets from current SAS drives from HGST or Seagate you find this:
4Kn: 4096, 4112, 4160, 4224
512e: 512, 520, 528)
So richard you said that T-10 DIF (Data Integrity Field) is used internally in the drive and below ZFS so that ZFS dont know anything about the extra bytes? For ZFS this drive is a normal 512 bytes drive? And with complementary you mean that the internal SAS drive error/checksum detection (DIF) can be used with ZFS? They dont interfere with each other? Thus i can use normal sas drives (which typically support 512/520/258 bytes) for any ZFS distributions (illumos,ZoL,FreeBSD,..)?
Thanks,
manuel.
PS: Another informative article regarding ZFS and SAS DIF: http://www.enterprisestorageforum.com/storage-technology/sas-vs.-sata-5.html
Gesendet: Freitag, 21. Oktober 2016 um 14:42 Uhr
Von: "Richard Elling" <[email protected]>
An: [email protected]
Cc: [email protected]
Betreff: Re: [developer] ZFS and SAS drives with 520 "fat" bytes
Von: "Richard Elling" <[email protected]>
An: [email protected]
Cc: [email protected]
Betreff: Re: [developer] ZFS and SAS drives with 520 "fat" bytes
On Oct 21, 2016, at 1:46 AM, [email protected] wrote:Hello,is it a good idea to use SAS drives with T10-DIF (520 Bytes) with ZFS (e.g. OmniOS or ZoL) ?Does ZFS support these drives or it is a bad idea to use these drives beacause ZFS handle its own checkums?
T10-DIF is a layer below ZFS. The two are complementary. The "fat bytes" are not exposed to ZFS.
T10-DIF only detects some failure modes, mostly those in-transit over fabrics. It won’t detect corruption
caused by overwrites, for example. To get true end-to-end detection you need checksums at the ends.
ZFS provides this for file system level. Clever applications provide this for themselves.
— richard
