Coda File System

Re: Coda development roadmap

From: <u-codalist-z149_at_aetey.se>
Date: Fri, 1 Aug 2014 13:22:08 +0200
On Thu, Jul 31, 2014 at 07:10:13PM -0500, Troy Benjegerdes wrote:
> Does this message end up in a maildir on a coda filesystem?
> 
> And if that is case, can you please point me to debian packages and 
> a possible migration path from afs to coda so I can try it?

There are Debian packages from the CMU team (dated 2010)
but they do not include the modular clog part.

My opinion is that it is a strenuous task to package according
to different ditro's preferences (many moving targets at once).

Coda makes also the distros much less necessary in general.
(We at Aetey are trying to make the world notice that it indeed does! :)

Of course there are Coda installers usable with Debian,
see the universal installer pages on the Coda wiki.

As for migration, I would be cautious when aiming to migrate from a
differing file system. Coda has nice rarely seen features but also
unusual kinds of limitations and failure modes.

> Also see the following for additional use cases:
> http://www.debian-administration.org/article/348/Distributed_filesystem_for_Debian_clusters

The symmetry is indeed the property which makes Coda highly attractice, it
allows for extremely nice modes of deployment and management - but beware
of expecting it to behave like a "'better' general purpose file system".

Both www.aetey.se and codawiki are happily running from Coda but this
does not mean it would fit your actual needs/expectations.
Your article does not tell how you want to be able to use the shared
file system - e.g. it would be not a good idea to place some opaque
database back end storage on Coda when the database engine expects a
local file system semantics.

> http://workshop.openafs.org/afsbpw06/talks/troy-afsworkshop-06.pdf

I have great empathy with your efforts and am working in basically the
same direction (but with a quite different approach).

This matter deserves certainly a longer and separate discussion.

In short, I see your experience like a try to blend two incompatible
world views. OS distros imply that everything is governed by the same
party as the host. Global systems are by their design independent of
host-specific details. These two contexts can not be used on each other's
premises without explicitly sacrificing the most useful virtues of both.
(Of course they _can_ be used together - it is only the traditional
practices which are highly inadequate in such a connection).

> Now imaging your mobile device, that boots an initrd that then mounts
> a debian (or ubuntu) based root filesystem hosted on Coda.

Not Coda nor any other really distributes system is a reasonable interface
for the properties normally assumed of a root file system in *nix.

> Now you have automatic synchronization of *everything* across all of
> your devices. Eventually a hardware vendor will figure out this is a
> killer app and deploy either OpenAFS or Coda.

One fundamental problem of distributed file systems is that it is very
hard to make money by contributing to them, at least to open source ones.
This stems probably from their global nature (how to lock-in the customer
when the technology opens the whole world?).

OpenAFS luckily survives, hope Coda also will find its ecological niche
(and a set of companies spending money on its development, Coda actually
has less complexity than AFS and would cost less ;)

Rune
Received on 2014-08-01 07:28:51