[Sheepdog] sheepdog question

MORITA Kazutaka morita.kazutaka at lab.ntt.co.jp
Mon Oct 10 19:28:49 CEST 2011


At Mon, 10 Oct 2011 13:56:18 +0200,
Mark Hunting wrote:
> 
> Sorry for not replying the original email, I don't have it as I only just subscribed to the list.
> 
> You recommend not using RAID, but from my point of view I have to disagree on this. For me data integrity is very important. The only way to achieve this with standard disks is by using a RAID controller with non-volatile cache. If plain disks are used without a RAID controller data integrity is not guaranteed, as the disk cache is volatile. This cache can be disabled of course, but that would be a terrible performance killer for sheepdog without a controller cache.
> When I think of it, I could configure a number of disks on my RAID controller as JBOD and run sheep daemons on these disks like you suggest. The disadvantage is that I will need a least 2 dedicated disks (RAID 1) for the OS then, and this makes the configuration less fool proof. And still, it's a requirement that all sheepdog disks are the same size on each server.
> 
> In an ideal world it would be possible to run a single sheep daemon on each server, with a different-sized backing devices on different servers. Sheepdog should then still make sure all data is always distributed across multiple servers.

Sheepdog ensures that data is not replicated in the same physical
machine by default.  But anyway, we have a plan to support different
sized disks and it is not so difficult, I think.

> Of course the result of this will be that some storage space may be lost if all servers have a different storage size (imagine two servers with 2 TB and one with 5 TB. Using --copies=3 this will result in only 2 TB space, the extra 3 TB on the 5 TB server will be unusable until at least a fourth server is added).
> 
> Besides reading this list now and then I didn't use sheepdog yet, and I'm not sure if the feature I described isn't already on the TODO. But I do use KVM a lot and Sheepdog looks very promising. I hope I can add a bit to the discussion without any hands-on sheepdog experience ;-) 

Thank you :)


Kazutaka

> 
> > Run multiple sheep daemons proportional to its capacity (3 daemons on
> > servers with 6x500 GB and one daemon on the other), and Sheepdog would
> > work as you expect, I think.  But the best way is to stop using RAID
> > and run a sheep daemon on for each disk.
> > 
> > Thanks,
> > 
> > Kazutaka
> 
> -- 
> sheepdog mailing list
> sheepdog at lists.wpkg.org
> http://lists.wpkg.org/mailman/listinfo/sheepdog



More information about the sheepdog mailing list