Re: [tlug] distributed file systems

[Attila Kinali <attila@example.com>]

On Mon, 15 Feb 2010 11:21:11 +0900
Sach Jobb <sach@example.com> wrote:

> We are setting up some local servers in the UK and and the US and I am
> thinking about implementing a distributed file system in order to
> solve data syncing problems. I first came across this looking at
> mogilefs (which appears to be perl) and then I have recently found a
> couple of fuse-based solutions: xtreemfs and hdfs. Does anyone have
> any opinions/recommendations/experiences on these (or others)?

>From the descriptions you gave here, i'd say coda is the filesystem
you want to use. At least it's the only one i've seen working
with non-local conections that is at least somewhat reliable even
on link-drop. And it has a cache build into the system.
But technically coda is old, very old. I have not tested the newer
distributed fs, so cannot compare them against coda.

But, considering my experience with distributed systems in generall
and how software in this area is designed and implemented, i strongly
suggest that you build your own synchronization system. If you know
a little bit about how networks on the global scale work and how
basic algorithms on distributed systems work, you can easily write
something that fits your needs exactly.

On the other hand, "pre-build" distributed systems are build on a
set of assumptions that might or might not fit your enviroment and
most of them are not documented anywhere. You might figure out some
of them while testing, but it's very likely that you will find the
one assumption that kills your application when you've been using
the system for months without problems... and then everything will
fall apart.


			Attila Kinali
-- 
If you want to walk fast, walk alone.
If you want to walk far, walk together.
		-- African proverb