[OpenAFS] Re: High Available Transparent Shared-Disk
Mon, 11 Apr 2011 10:08:34 -0500
On Sun, 10 Apr 2011 03:45:54 -0700 (PDT)
a a <email@example.com> wrote:
> Hi All,
Please don't multipost.
> We are searching around a solution like a "high available transparent
> file system" that makes the fault transparent to the application, so
> in case of fault, redundant machine still can access the files even
> the master machine is down (replica issue or such a thing).
> It seems OpenASF doesn't have fail-over feature, am I right? My
> question is that can OpenASF help us in our case?
What is typically said is that, out of the box, OpenAFS has redundany
for read-only data. What this means is that you can write data to one
location (the "read-write volume"), and later on you decide when to
distribute that data to multiple read-only sites ("releasing the
volume"). After that, clients can read the data from the read-only
sites, and will failover automatically if any of those sites fail.
Releasing the volume data is normally done as an administrative action.
It is automatable, and certainly some sites do automated releases a lot,
but it's probably not useful to do it very frequently (e.g. every 5
seconds). So OpenAFS doesn't really have a way to replicate data as it
is written, right now. However, read-write replication is a known
desired feature already, and some work has been done in that area, so it
may exist in the future.
However, it is still possible today to achieve some kind of read-write
replication with OpenAFS volumes by just using some other kind of
underlying storage. If you create your AFS volumes on a SAN, or
DRBD-backed storage, you can achieve redundancy for read-write data if
you bring up a fileserver on another machine when an outage occurs.
Many sites have AFS data on a SAN that they can quickly serve from
different fileservers, and I know of at least one site that has
DRBD-backed fileservers that I believe have worked well.
Also, since you seem to be asking this same question of several projects
(the exact same question, in fact :), you may want to look at the Coda
filesystem. Coda is very AFS-like, and offers built-in read-write
replication. I believe it is not as mature/stable as OpenAFS, but you
can judge for yourself if it fits your needs.