On Sun, 17 Mar 2019 09:15:52 +0900, Raymond Burkholder wrote: > > On 2019-03-15 11:29 p.m., MORITA Kazutaka wrote: > >> Any suggestions on alternatives? I'd say ceph is a primary one, but > >> sheepdog seemed to be 'lighter weight'. > > > > I'm a bit interested in keeping sheepdog as a tiny implementation of > > distributed block storage so that we can use it for evaluation purposes. > > Since we don't have enough resources for maintenance, we > > have to remove some of the current sheepdog features and make > > implementation as simple as possible. For example: > > - Remove object cache feature. We can create a block-level cache > > outside of sheepdog with bcache or dm-cache. > > - Remove hyper volume feature. We can create a huge block device on > > top of sheepdog volumes with LVM. > > - Remove object storage feature. It's out of scope of distributed > > block storage and there is more stable software like OpenStack Swift. > > - Remove experimental feature like NFS or Linux block device support. > > > > Any comments would be welcome. > > > I would imagine that does mean that you would be keeping the block storage > which KVM knows how to use? Which I think was a primary goal? That was my > primary reason I was giving Sheepdog a try. > > I'm not sure which specific function above to which this relates, but the > other > interesting feature was the shared file storage. Being able to share files > between hosts was of value. But I could see that being removed if a simple > distributed file system is too complex.
I thought of keeping only a block device interface for KVM, and shared file storages like NFS was out of scope. However, it might be better to refine Sheepdog as a tiny library (libsheepdog) to implement distributed storage system, and provide a sample implementation of it as a block storage for KVM. Then, someone has a chance to create a shared file storage on top of sheepdog. Thanks, Kazutaka -- sheepdog mailing list sheepdog@lists.wpkg.org https://lists.wpkg.org/mailman/listinfo/sheepdog
