On 12/28/2011 09:09 PM, Christoph Hellwig wrote: > On Sun, Dec 25, 2011 at 11:42:46PM +0800, Liu Yuan wrote: >> - auto checksumed snapshot objects > > Did you ever look into using the auto compression/checksum feature also > for regular data objects? While it might not be optimal for some > workload it seems like a useful feature when we might want automatic > data de-duplication and/or have a read mostly workload. A typical > example would be base images for the OS disk which are shared by a lot > of VMs and might even get the same kind of modifications to them. > Let's stay with Farm's working directory for regular data operations, in order to gain comparable I/O performance, suitable for general workloads. I think LevelDB will match what you desire, they do tricks optionally when they 'put' those objects into persistent storage. I am having it in mind to adopt LevelDB as Sheepdog's backend store later, (not a promise, though I guess it's just several hundred lines to implement a LevelDB driver). ;) Thanks, Yuan |