[Sheepdog] Sheep gateway question

[Matthew Law]

On Wed, November 9, 2011 3:10 am, MORITA Kazutaka wrote:
> Different VMs can use different gateways, so if there are many VMs,
> the gateway cannot be a bottleneck.  Our goal is not to provide the
> combined bandwidth of many disks to one VM, but rather to achieve
> scalability; Sheepdog provides enough bandwidth to each VM even if
> many VMs use Sheepdog at the same time.

How is the gateway node decided?  Perhaps we could add this to the
documentation to explain to anyone who jumped to the same (wrong!)
conclusion as me? :-)

I think for my use case and for the majority of people 'enough bandwidth'
is simply that the VMs do not 'feel' slow when reading and writing and
that I/O is predictable.  Nobody can expect clustered storage of this type
to perform like a high performance disk array (this is not the use case). 
If it scales well, performs predictably and I/O latency is acceptable then
I think it will appeal to many.

Perhaps the sheepdog home page could add a section something like this:

You should consider Sheepdog if you are looking for clustered storage that:

  * Is 'scale-out', with no single point of failure
  * Is simple to setup and use (secured at the network layer)
  * Is compatible with KVM via QEmu (Xen support is coming later...)
  * Works at the block-level
  * ...

You should not consider Sheepdog if you are looking for:

  * High bandwidth/low latency storage ('scale-up')
  * Fine grained authentication and authorization of storage nodes
  * Works at the file level (clustered file system)
  * Tiering
  * ...

Just a suggestion.

Thanks,

Matt.