Re: [Gluster-devel] Hello... stripping and fault tolerant systems...

[Kevan Benson]

Victor San Pedro wrote:
> Hello. I would like to ask you the following question:
>
> Is it possible to use cluster STRIPING to build a fault tolerant system?
>
> For example:
>
> If I have 6 storage servers
> I would like to storage my files, all have a big size,  splited in the 6
> servers ( I think I can achieve that with the striping cluster)
> The goal would be to build a fault tolerant system with this striping
> cluster with 6 servers...
>
> In the case a server fail, I could still access to all my files...
> Could I managed this aim joining STRIPING with AFR? It is possible to do
> that with clusterfs (it seems not to be model spec files of it)?
>
> How could I managed this aim? Any ideas?
> What is your opinion about other possible ways of managing a fault
> tolerant system using as less space as possible with glusterfs?

If you export two volumes per server, and chain your AFR subvolumes 
between servers, you should get a fault tolerant system of the type you 
are looking for.

Something like this:

Brick X exports two volumes, brickX-1 and brickX-2

UNIFY {
  AFR1 { brick1-2, brick2-1 }
  AFR2 { brick2-2, brick3-1 }
  AFR3 { brick3-2, brick4-1 }
  AFR4 { brick4-2, brick5-1 }
  AFR5 { brick5-2, brick6-1 }
  AFR6 { brick6-2, brick1-1 }
}

If you wanted MORE redundancy, you could export three volumes per 
server, and make you AFR's like this:
UNIFY {
  AFR1 { brick1-3, brick2-2, brick3-1 }
  AFR2 { brick2-3, brick3-2, brick4-1 }
...

Which would be able to survive two server failures without data loss.

--

-Kevan Benson
-A-1 Networks