Hello Pierre,
I see your point and I understand your arguments. But as it happens, the
way you create your volumes has an impact on what you can do afterward,
and how you can expand them. *It has nothing to do with GlusterFS
itself, nor the developers or the community, it's all in your
architectural choices in the beginning.*
It would be the same problem with many other distributed FS, or even
some RAID setups. That is the kind of thing that you have to think about
in the very beginning, and that will have consequences later.
Aytac's argument doesn't apply to you as you don't have a replicated
volume, but a distributed+striped one. So if you lose one of your
original 4 nodes, you lose half of one of your stripes, and that will
make the files on that stripe inaccessible anyway. If you lose a node
with both bricks in the stripe, well the result will be the same. So I
don't think that the colocation of striped bricks on the same host
matters much in your case, outside of performance concerns.
As for your 14-node cluster, if it's again a distributed+striped volume,
then it's a 2 ? 7 volume (distributed over 2 brick groups, each being a
7-disk stripe). To extend it, you will have to add 7 more bricks to
create a new striped brick group, and your volume will be transformed
into a 3 ? 7 one.
I would advise you to read carefully the master documentation pertaining
to volume creation and architecture. Maybe it will help you understand
better the way things work and the impacts of your choices:
https://github.com/gluster/glusterfs/blob/master/doc/admin-guide/en-US/markdown/admin_setting_volumes.md
Thanks,
JF
On 17/03/15 10:20, Pierre L?onard wrote:> Hi aytac zeren,
>> In order to have a healthy volume, you need to have bricks as factors
>> of 4. Which means if you have 4 bricks and want to extend your setup
>> with additional brick, then you will need 4 other bricks in your
>> cluster, for your setup.
>>
>> I don't recommend to host more than one brick on a host as it would
>> cause data loss on failure of the node, if your master and redundant
>> copy is stored on the same host.
>> Regards
>> Aytac
> Yes I understand. But You can understand and I hope that the gluster
> team can understand, that if I want to expand my computational power and
> storage I can't buy four node In one time, it's to heavy.
> Another example, I have a big cluster with 14 node stripe 7. I can't by
> 14 other nodes to expand it in one time. It's a big limitation in the
> flexible usage of glusterfs.
>
> Does that mean that glusterfs is only dedicated to small cluster ?
>
> Sincerely.
> --
> Signature electronique
> INRA <http://www.inra.fr>
>
> *Pierre L?onard*
> *Senior IT Manager*
> *MetaGenoPolis*
> Pierre.Leonard at jouy.inra.fr <mailto:Pierre.Leonard at
jouy.inra.fr>
> T?l. : +33 (0)1 34 65 29 78
>
> Centre de recherche INRA
> Domaine de Vilvert ? B?t. 325 R+1
> 78 352 Jouy-en-Josas CEDEX
> France
> www.mgps.eu <http://www.mgps.eu>
>
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
>
--
Jean-Fran?ois Le Fill?tre
-------------------------------
HPC Systems Administrator
LCSB - University of Luxembourg
-------------------------------
PGP KeyID 0x134657C6