zfs discuss - Feb 2010 - Oracle Performance - ZFS vs UFS (Jason King)

> There is of course the caveat of using raw devices with databases (it
> becomes harder to track usage, especially as the number of LUNs
> increases, slightly less visibility into their usage statistics at the
> OS level ).   However perhaps now someone can implement the CR I filed
> a long time ago to add ASM support to libfstyp.so that would allow
> zfs, mkfs, format, etc. to identify ASM volumes =)
While that would be nice, I would submit that if using ASM, usage becomes solely
a DBA problem.  From the OS level, as a system admin, I don''t really
care?I refer any questions back to the DBA.  They should have tools to deal with
all that.

OTOH, with more things stacked on more servers (zones, etc.) I might care if
there''s a chance of whatever Oracle is doing affecting performance
elsewhere.

Thoughts?

My problem is when you have 100+ luns divided between OS and DB,
keeping track of what''s for what can become problematic.   It becomes
even worse when you start adding luns -- the chance of accidentally
grabbing a DB lun instead of one of the new ones is non-trivial (then
there''s also the chance that your storage guy might make a mistake and
give you luns already mapped elsewhere on accident -- which I have
seen happen before).  And when you''re forced to do it at 3am after
already working 12 hours that day.... well safeguards are a good
thing.


On Sat, Feb 13, 2010 at 2:13 PM, Allen Eastwood <mixal at paconet.us>
wrote:
>
>> There is of course the caveat of using raw devices with databases (it
>> becomes harder to track usage, especially as the number of LUNs
>> increases, slightly less visibility into their usage statistics at the
>> OS level ). ? However perhaps now someone can implement the CR I filed
>> a long time ago to add ASM support to libfstyp.so that would allow
>> zfs, mkfs, format, etc. to identify ASM volumes =)
>
> While that would be nice, I would submit that if using ASM, usage becomes
solely a DBA problem. ?From the OS level, as a system admin, I don''t
really care?I refer any questions back to the DBA. ?They should have tools to
deal with all that.
>
> OTOH, with more things stacked on more servers (zones, etc.) I might care
if there''s a chance of whatever Oracle is doing affecting performance
elsewhere.
>
> Thoughts?
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
>

So, one of the tricks I''ve used in the past is to assign a volname in
format as I use luns.  Dunno if that''s an option with ASM?  ZFS seems
to blow those away, the last time I looked.

-A

On Feb 13, 2010, at 14:32 , Jason King wrote:
> My problem is when you have 100+ luns divided between OS and DB,
> keeping track of what''s for what can become problematic.   It
becomes
> even worse when you start adding luns -- the chance of accidentally
> grabbing a DB lun instead of one of the new ones is non-trivial (then
> there''s also the chance that your storage guy might make a mistake
and
> give you luns already mapped elsewhere on accident -- which I have
> seen happen before).  And when you''re forced to do it at 3am after
> already working 12 hours that day.... well safeguards are a good
> thing.
> 
> 
> On Sat, Feb 13, 2010 at 2:13 PM, Allen Eastwood <mixal at paconet.us>
wrote:
>> 
>>> There is of course the caveat of using raw devices with databases
(it
>>> becomes harder to track usage, especially as the number of LUNs
>>> increases, slightly less visibility into their usage statistics at
the
>>> OS level ).   However perhaps now someone can implement the CR I
filed
>>> a long time ago to add ASM support to libfstyp.so that would allow
>>> zfs, mkfs, format, etc. to identify ASM volumes =)
>> 
>> While that would be nice, I would submit that if using ASM, usage
becomes solely a DBA problem.  From the OS level, as a system admin, I
don''t really care?I refer any questions back to the DBA.  They should
have tools to deal with all that.
>> 
>> OTOH, with more things stacked on more servers (zones, etc.) I might
care if there''s a chance of whatever Oracle is doing affecting
performance elsewhere.
>> 
>> Thoughts?
>> _______________________________________________
>> zfs-discuss mailing list
>> zfs-discuss at opensolaris.org
>> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
>>