Gluster users - Jun 2016 - stripe 2 replica 2 VS disperse 4 redundancy 2

Hi:

I have a big doubt. 
I have 2 servers with 2 disks of 2 TB each. I've been thinking to create a
volume with stripe 2 replica 2 creating a brick with each disk and using
server1:/b1 server2:/b1 server1:/b2 server2:/b2.
This seems to work fine 4TB of space and if one disk or even one server fails
the volume is still there. But I just found disperse volumes, I understand that
disperse 4 redundancy 2 work in the same way.

Any suggestion on which solution is better? which one is faster? which one
you'll recommend?

Thanks from Canary Islands

Manuel Padr?n Mart?nez

On 06/17/2016 03:05 AM, Manuel Padr?n Mart?nez wrote:
> Hi:
>
> I have a big doubt.
> I have 2 servers with 2 disks of 2 TB each. I've been thinking to
create a volume with stripe 2 replica 2 creating a brick with each disk and
using server1:/b1 server2:/b1 server1:/b2 server2:/b2.
Striping is not actively developed. Sharding [1] is the successor to
it.
> This seems to work fine 4TB of space and if one disk or even one server
fails the volume is still there. But I just found disperse volumes, I understand
that disperse 4 redundancy 2 work in the same way.
>
> Any suggestion on which solution is better?
It depends on your workload really.
>   which one is faster?
Replica volumes are faster than disperse because there is no erasure 
code math to be done during I/O but  as is obvious, you'd get less 
volume space than disperse.
> which one you'll recommend?
For  high I/O rate workloads, replica could be a better choice. You 
should try both and see what works best for you.
Btw, you need 6 bricks for a 4+2 disperse configuration.

-Ravi

[1] http://blog.gluster.org/2015/12/introducing-shard-translator/
>
> Thanks from Canary Islands
>
> Manuel Padr?n Mart?nez
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users

Hola Manuel,

as Ravishankar has said, you should use sharding instead of stripe.

Regarding the disperse, the minimum number of servers you would need is 
3. Disperse requires at least 3 bricks to create a configuration with 
redundancy 1 (this is equivalent to a replica 2 in terms of redundancy), 
but if you put 2 of those bricks in a single server and that server 
dies, you will lose 2 bricks from a volume that only tolerates 1 brick 
failure.

You can have 2 or more bricks per server and create a 
distributed-disperse setup.

The "disperse 4" means that each disperse-set is made of 4 bricks. 
"redundancy 2" means that two of those bricks will be redundant (i.e.
it
will work fine with 2 bricks down). However disperse doesn't allow 
redundancies greater or equal to "number of bricks" / 2.

With redundancy 2, the minimum number of bricks/servers should be 5.

For better performance in most cases, it's recommended to have a 
configuration where "number of bricks" - "redundancy bricks"
is a power
of 2 (i.e. 2, 4, 8 or 16).

Currently, replica is a bit faster than disperse, however with disperse 
you waste less physical storage for the same usable space.

If you have enough servers, you should try it and see if the performance 
is good enough for your workload.

Xavi

On 16/06/16 23:35, Manuel Padr?n Mart?nez wrote:
> Hi:
>
> I have a big doubt.
> I have 2 servers with 2 disks of 2 TB each. I've been thinking to
create a volume with stripe 2 replica 2 creating a brick with each disk and
using server1:/b1 server2:/b1 server1:/b2 server2:/b2.
> This seems to work fine 4TB of space and if one disk or even one server
fails the volume is still there. But I just found disperse volumes, I understand
that disperse 4 redundancy 2 work in the same way.
>
> Any suggestion on which solution is better? which one is faster? which one
you'll recommend?
>
> Thanks from Canary Islands
>
> Manuel Padr?n Mart?nez
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
>