RFC: High availability replication/raid filesystem



Hi,

With regards to high availability and hosting of email & websites, I
was thinking of a FUSE filesystem which provides a transparent layer
to any current filesystem and provides replication and raid-like
behavior for files stored on it. Something very very simple, using the
normal directory structure and and its own little .whatever directory
for its own info ... this way if everything goes to hell, stop fuse and
use the FS normally.

Busy throwing stones in bushes here to see if this is truely a good
idea, but ... think for a second of a network of servers configured in
different ways, for instance 2 masters with replicating filesystems
posting changes between eachother, 5 or so slaves around the world
providing geographically different access points to the data. When a
master falls off the net (voted to be removed by all the other nodes if
it cannot be reached), another "master" can be elected. When the one
that was down comes backup, it will replay the transactions the new
master has logged, ask for a vote and regain its status.

Now also think about 5 masters, with raid-like behaviour .... peices of
even some files can be stored on either one and rebuilt if a master
becomes unavailable on the fly from the rest of the data. This can
spread data geographically without the need for one centralized storage
area.

The power of FUSE lies in its userspace architecture, think about
replication of files which only match a regex, or even raid'ing of
files which match a regex ... the possibilities are limitless! It may
be userspace ... but I'm very sure capable dealing with the demands of
a high availability application.

Its just a thought, but with the amount of mail and web my organization
does and the need for more redundancy on both of these platforms I'm
seriously considering sourceforging this project (GPL) and starting
work asap.

Anyone interested or want to play devils advotate?

Regards
Nigel

.



Relevant Pages