We have been using Lustre for over five years, and had experienced and learned a
lot from deploying, debugging and modifying it. For us, Lustre is powerful and
amazing.
Recently, we are implementing a high availability system for Lustre, and are
looking forward to open-sourcing it. Basically, it is a stackable file system
which uses Lustre or other file system''s directories as its lower
storage, and automatically generate multiple identical branches of files or
directories, while everything being keeped transparent to users. Files or
directories with some broken branches can be automatically healed as long as one
of their branches is fine.
We used to use Heartbeat for automatic failover, but we think it would be a
good ideal to keep duplications of data too. Sometimes disk arrays just break
down and datas are lost forever. And it is why we design and implement this
system.
Our system is under heavy development and may be stable for open-sourcing in
a few months. We are open for any ideal from its basic designs to its
implementation details. Is there anyone interested in this project? Any
suggestion or question will be welcome.
Thank you!