zfs discuss - Aug 2006 - Tunable parameter to zfs memory use

Hi

been using now zfs since 06/06 u2 release has been out.
one thing have notised.
zfs eats a lot of memory.
right after boot mem usage is about 280M
but after accessing zfs disks usage rises fast to be 900M.
and it seems to stay in level of 90% of tot mem.
also noted it frees used mem but running "heavy" apps you see
performace impact while zfs frees memory to other use.

now idea:
could there be /etc/system tunable parameter for zfs so we could manually
set % of total memory zfs can use.

like :
zfs_max_phys_mem 30%
would tell zfs that it can use 30% of available memory.
and via versa
zfs_min_phys_mem 10%
would tell zfs that it can allways use 10% of available memory.
--> these maybe need to global values for all zfs filesystems 

or just simple zfs option that can be set as other zfs options , like
compression, etc... then this setting could be easily per zfs filesystem

Thanks
 
 
This message posted from opensolaris.org

homerun wrote:
> been using now zfs since 06/06 u2 release has been out. one thing
> have notised. zfs eats a lot of memory. right after boot mem usage is
> about 280M but after accessing zfs disks usage rises fast to be 900M.
>  and it seems to stay in level of 90% of tot mem. also noted it frees
> used mem but running "heavy" apps you see performace impact while
zfs
> frees memory to other use.
Yes, that is the case. As I understand things it''s an artefact of
how zfs is implemented. There are changes in the Solaris 10 patch
pipeline which should make the kmem usage somewhat better.
> now idea: could there be /etc/system tunable parameter for zfs so we
> could manually set % of total memory zfs can use.
> like : zfs_max_phys_mem 30% would tell zfs that it can use 30% of
> available memory. and via versa zfs_min_phys_mem 10% would tell zfs
> that it can allways use 10% of available memory. --> these maybe need
> to global values for all zfs filesystems
> or just simple zfs option that can be set as other zfs options , like
> compression, etc... then this setting could be easily per zfs
> filesystem
I understand the concern, but I''m not sure that this is the
best way of achieving your goal.

What would you want to observe if your system hit the upper
limit in zfs_max_phys_mem?


cheers,
James C. McPherson

> What would you want to observe if your system hit the upper
> limit in zfs_max_phys_mem?
I would want zfs to behave well and safely like every other app on which you
apply boundary conditions.  It is the responsibility of zfs to know its
boundaries and stay within them.  Otherwise, your system may exist only for zfs
and not for the application or services you wish to run on that system of which
zfs is just one part.


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