Hi list, I just learned about Stratis, and it is a very exciting project! The idea of ZFS-like features, while preserving proper layering in the I/O stack, in a single, simple package is really cool.
I read the Stratis PDF documentation and, if I understand it properly, you plan to use dm-integrity to checksum blocks on the I/O device. I wonder if you considered extending current MDRAID code to integrate checksumming capability: in this manner, you not only "shave off" a I/O layer (no need of dm-integrity on the single block devices), but *any* new MD array become automatically checksum-protected.
Some years ago, this very idea was heavily debated on MDRAID mailing list. At the time BTRFS looked "almost ready", so the MDADM maintainer decided that disk's own hardware ECC was enough to protect from the most likely problems. Maybe this can be right time to revisit this position?
Thanks.
On 08/15/2017 10:54 AM, Gionatan Danti wrote:
Hi list, I just learned about Stratis, and it is a very exciting project! The idea of ZFS-like features, while preserving proper layering in the I/O stack, in a single, simple package is really cool.
I read the Stratis PDF documentation and, if I understand it properly, you plan to use dm-integrity to checksum blocks on the I/O device. I wonder if you considered extending current MDRAID code to integrate checksumming capability: in this manner, you not only "shave off" a I/O layer (no need of dm-integrity on the single block devices), but *any* new MD array become automatically checksum-protected.
Some years ago, this very idea was heavily debated on MDRAID mailing list. At the time BTRFS looked "almost ready", so the MDADM maintainer decided that disk's own hardware ECC was enough to protect from the most likely problems. Maybe this can be right time to revisit this position?
Hi Gionatan,
I think we would use it if it was available upstream, but I don't know if championing its adoption is a good use of our time right now. We just started working on the 0.5 milestone -- focused on snapshots, cache tier, logging, and DBus improvements. RAID support isn't planned until 2.0.
-- Andy
Il 15-08-2017 20:47 Andy Grover ha scritto:
Hi Gionatan,
I think we would use it if it was available upstream, but I don't know if championing its adoption is a good use of our time right now. We just started working on the 0.5 milestone -- focused on snapshots, cache tier, logging, and DBus improvements. RAID support isn't planned until 2.0.
-- Andy
Hi Andy, I understand. For what I know, checksumming is *not* present on current upstream MDRAID (see here for a recent discussion about bit-rotting: http://marc.info/?l=linux-raid&m=148857638032570&w=2)
Go ahead with the good work :) Regards.
stratis-devel@lists.stg.fedorahosted.org