[illumos-Developer] illumos bug/rfe 453

[wessels]

Hi,

indeed zfs isn't relevant here, I meant zpool. And yes zpool can
handle larger sector sizes already. The size is stored in the ashift
property upon creation. Patched zpool versions do exist which override
ashift and fix it to 12bits (4K). I've read the various topics and
benchmarks regarding the WD20EARS for example.
I've also read the reports of succesfull use of mixed sector sizes in
zpool's. But I didn't study the code so don't know for sure if the
zpool programmers had the intention to support different sector sizes
in one zpool. That's why I mentioned the cases concerning mixed
devices, eg harddisks 512bytes and L2ARC ssd with reported 512bytes
but would benefit greatly when aligned 4K i/o's would be done against
it.
Are you aware of any device SAS or SATA which actually reports 4K
sectors? That would make testing more interesting. I was hoping that
the new hitachi drives would report 4K, but they don't.
What format do you suggest for the SD tunable list ? Something in the
form of a list of id's (vendor, model, firmware etc) and per model a
requested size? A different scheme would be per device (port). But
that require much more maintenance.
Regarding the test suite. That would be very interesting. A SD driver
reporting 4K and a test suite which check's if the data is nicely
aligned on this 512bytes disk. I wanted to check this once on a
windows system and had to boot some linux toolbox to read the disk
realiable.
If you want I can lookup when the other two arc's were integrated. I
don't have the new Solaris 11 SXCE or whatever it's called which would
be interesting to use as reference for PSARC/2010/296

Regards