On 02/14/2012 01:25 PM, John Mark Walker wrote: > ----- Original Message ----- >> I'm currently fighting to get GlusterFS replica in an HPC environment >> but the "wasting half the space" argument is hard to fight when >> there's >> a tight budget. There really is no waste at all, the space is being >> used for full server redundancy (IMHO you need server redundancy, not >> just disk redundancy) and in some use-cases, increased performance >> (in >> other use-cases replica is slower). > > I think this gets to the heart of the matter. This is very much on our minds as we look at future roadmaps. > > For now, replica 2 + RAID underneath is a valid solution for the vast majority of use cases. True, but it seems like at some point the cost of N+N just looks silly (think if you need 2,000 servers to get the usable space of 1,000 of them). > There shouldn't be many cases where replica 3 is absolutely necessary - not when you utilize a decent RAID card. > To pose the issue in a slightly different light, what would you want future behavior to be? Is erasure coding something that you view as essential in the near future? > > -JM By "erasure coding" I assume you mean "some RAID 5/6-like data recovery with parity". I think this should be something to investigate and see if it is plausible to build. At the very least it opens GlusterFS to more use-cases and more design choices.
