[sheepdog-users] Multiple disks per server

[Matthew Law]

On Mon, July 2, 2012 5:58 am, Alexandre DERUMIER wrote:
> Hi,
>
> Is sheepdog aware of the location of sheepdog daemons ?

Yes it is.

> Can the replication occur on sheepdog daemons on same server ?

Yes it can and it works very well indeed.

I have been testing it on three servers which have 4 disks each and boot
them from PXE to give all storage to sheepdog. Each sheep then gets an
entire disk for storage.

You can specify which nodes the disks are attached to so data is still
properly distributed and there is no danger of data loss should a node go
down and take multiple disks with it.

The converse is true however and you can, for example, build a local RAID
array and give that to a single sheep if you so wish. My personal
preference is to run one sheep per disk and get as much bandwidth and IOPS
as you can from the disks you have.

I am doing more testing and then hoping to document how to build a PXE
booting cluster with all disks dedicated to sheep but £work hasn't let me
get that far yet.  Hopefully soon!


Cheers,

Matt.