Erasure Coded backend?
Harry Mangalam
BeeGFS mentioned that it was going to support an erasure coding back-end, now available in some other distributed and object filesystems. This would be a pretty dramatic change in storage and would allow (for some applications) reducing $ on backup hardware.
Could a ThinkParq representative comment on when this feature is going to be released? Is this in Release 6? 6.X? months? etc?
I'm assuming that this would have to be a fresh installation; there would be no ability to transition to such a back-end. Is that right?
And does this require any special-purpose hardware or setup such as dual-path controllers?
We are planning a major (for us) campus storage pool and this could affect decision-making.. This was one of the few things that prevented BeeGFS from being considered for this project (metadata mirroring was another, but that is supported now).
Also, since I'm asking for everything :) is there any thought to forking or modifying the RobinHood Policy Engine to talk directly to the metadata server like it does to Lustre? With even small cluster filesystems often in the multi-PB range, this kind of real-time oversight is becoming more critical since external FS scans can now take days to complete.
hjm
