zfs discuss - Aug 2008 - Disabling disks'' write-cache in J4200 with ZFS?

I
want to disable write cache on the disk drives in our J4200 JBODs so
that fsync() actually writes to disk, not just to the cache on the
drive.

I did
this using ''format -e'', but it displays a warning about the
drive being
part of a zpool and also it says that the change is not permanent.

Is
''format -e'' the right way to do this with ZFS?  Is there no
way to make
it permanent? 

-- 
Todd E. Moore
Sun Microsystems Incorporated
443.516.4002
AIM: toddmoore72462


_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

Todd E. Moore wrote:
>
> I want to disable write cache on the disk drives in our J4200 JBODs so 
> that fsync() actually writes to disk, not just to the cache on the drive.
>
ZFS will do this for you, via the way the ZIL works.
Neil explains it pretty well at
http://blogs.sun.com/perrin/entry/the_lumberjack
> I did this using ''format -e'', but it displays a warning
about the
> drive being part of a zpool and also it says that the change is not 
> permanent.
>  
> Is ''format -e'' the right way to do this with ZFS?  Is
there no way to
> make it permanent?
In general, you don''t need to do this.  ZFS will send the cache
flush command to the disks as needed.
 -- richard

I''m working with a group that wants to commit all
the way to disk every single write - flushing or bypassing all the
caches each time.  The fsync() call will flush the ZIL.  As for the
disk''s cache, if given the entire disk, ZFS enables its cache by
default.  Rather than ZFS having to issue the flush command to the disk
we want to disable this cache and avoid the step altogether

Richard Elling wrote:
Todd E.
Moore wrote:

I want to disable write cache on the disk drives in our J4200 JBODs so
that fsync() actually writes to disk, not just to the cache on the
drive.

ZFS will do this for you, via the way the ZIL works.

Neil explains it pretty well at

http://blogs.sun.com/perrin/entry/the_lumberjack

  I did this using ''format -e'', but it displays
a warning about the drive being part of a zpool and also it says that
the change is not permanent.

Is ''format -e'' the right way to do this with ZFS?  Is there no
way to
make it permanent?

In general, you don''t need to do this.  ZFS will send the cache

flush command to the disks as needed.

-- richard

-- 
Todd E. Moore
Sun Microsystems Incorporated
443.516.4002
AIM: toddmoore72462


_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

Todd E. Moore wrote:
> I''m working with a group that wants to commit all the way to disk
every
> single write - flushing or bypassing all the caches each time. The
> fsync() call will flush the ZIL. As for the disk''s cache, if given
the
> entire disk, ZFS enables its cache by default. Rather than ZFS having to
> issue the flush command to the disk we want to disable this cache and
> avoid the step altogether
Then tell your (in my opinion insane) group to pass O_SYNC to open(). 
That will guarantee writes go to disk without being cached (except by 
non-volatile cache in a raid controller). If you don''t want the ZIL 
involved, don''t configure one for your storage.

-- 
Carson

Carson Gaspar wrote:
> Todd E. Moore wrote:
>> I''m working with a group that wants to commit all the way to
disk every
>> single write - flushing or bypassing all the caches each time. The
>> fsync() call will flush the ZIL. As for the disk''s cache, if
given the
>> entire disk, ZFS enables its cache by default. Rather than ZFS having
to
>> issue the flush command to the disk we want to disable this cache and
>> avoid the step altogether
> 
> Then tell your (in my opinion insane) group to pass O_SYNC to open().
> That will guarantee writes go to disk without being cached (except by 
> non-volatile cache in a raid controller). If you don''t want the
ZIL
> involved, don''t configure one for your storage.
I think there''s some confusion here. For a description of the ZIL
and separate log devices see:

http://blogs.sun.com/perrin/entry/the_lumberjack
http://blogs.sun.com/perrin/entry/slog_blog_or_blogging_on

If you don''t configure a separate log device - I assume this
is what you meant by "ZIL" above), then the intent log is
embedded in the main pool. It''s not advisable to disable the
ZIL code as this actually handles fsync/O_[D]SYNC.
Without it those synchronous requests are ignored,
which is the opposite of what Todd wants.

Neil.

Todd E. Moore wrote:
> I''m working with a group that wants to commit all the way to disk 
> every single write - flushing or bypassing all the caches each time.  
> The fsync() call will flush the ZIL.  As for the disk''s cache, if 
> given the entire disk, ZFS enables its cache by default.  Rather than 
> ZFS having to issue the flush command to the disk we want to disable 
> this cache and avoid the step altogether
Actually, it will flush to the ZIL.

If you don''t want the disk''s cache enabled by default, then
use a slice.  None of this is difficult.
 -- richard