Gluster users - Jun 2008 - balancing redundancy with space utilization

Currently it would seem that AFR will simply copy everything to every 
brick in the AFR.  If I did something like ...

volume afr-example
   type cluster/afr
   subvolumes brick1 brick2 brick3 brick4 brick5 brick6 brick7 brick8
end-volume

I would wind up with 8 copies of every file.  Clearly, this is too many. 
  What I would rather have is maybe 3 copies of each file distributed 
randomly across 3 servers, so that I could still have 2 servers fail and 
  have all data available, but without using up unnecessary space on the 
other 5.  The 3 would need to be round-robined in some manner so as to 
distribute the disk utilization.  First file goes on brick1 brick2 
brick3, 2nd file goes on brick2 brick3 brick4, etc.

It seems that AFR used to have this with "option relicate *:3" but
that
was removed.  The supposed replacement for that, the switch scheduler, 
doesn't really have the same functionality.

Unless there is a undocumented form of the "option switch.case" 
statement that I have yet to see.  Can I do "option switch.case *:3"
or
some such?

-- 
-- ===========================    Tom Lahti
    BIT Statement LLC

    http://www.bitstatement.net/
-- ============================

Hi Tom,
You need unify of afr volumes.

3 copies of 8 servers is slightly odd to pair. Though you
have options..
1) 1-2-3, 4-5-6, 7-8-1
2) 1-2-3, 4-5-6, 7-8
3) 1-2-3, 2-3-4, 3-4-5 4-5-6 5-6-7 6-7-8

My recommendation is to go for 9 servers if you are looking
for 3 copies of all files. Upgrading in pairs of 3 becomes
easier at a later stage.

Please refer to this documentation:
http://www.gluster.org/docs/index.php/Unify_over_AFR

Happy Hacking,
--
Anand Babu Periasamy
GPG Key ID: 0x62E15A31
Blog [http://ab.freeshell.org]
The GNU Operating System [http://www.gnu.org]
Z RESEARCH Inc [http://www.zresearch.com]



Tom Lahti wrote:
> Currently it would seem that AFR will simply copy everything to every 
> brick in the AFR.  If I did something like ...
> 
> volume afr-example
>    type cluster/afr
>    subvolumes brick1 brick2 brick3 brick4 brick5 brick6 brick7 brick8
> end-volume
> 
> I would wind up with 8 copies of every file.  Clearly, this is too many. 
>   What I would rather have is maybe 3 copies of each file distributed 
> randomly across 3 servers, so that I could still have 2 servers fail and 
>   have all data available, but without using up unnecessary space on the 
> other 5.  The 3 would need to be round-robined in some manner so as to 
> distribute the disk utilization.  First file goes on brick1 brick2 
> brick3, 2nd file goes on brick2 brick3 brick4, etc.
> 
> It seems that AFR used to have this with "option relicate *:3"
but that
> was removed.  The supposed replacement for that, the switch scheduler, 
> doesn't really have the same functionality.
> 
> Unless there is a undocumented form of the "option switch.case" 
> statement that I have yet to see.  Can I do "option switch.case
*:3" or
> some such?
>