[HCoop-Discuss] Reorganizing, people-wise and tech-wise

[Matthias-Christian Ott]

On Fri, Jun 26, 2009 at 07:47:26AM -0400, Adam Chlipala wrote:
> Matthias-Christian Ott wrote:
> >
> >>> I vote against separate machines, my experience at TIP9UG showed that
> >>> it's quite valuable to have a shared filesystem.
> >>>   
> >>>       
> >> Can you elaborate on this?
> >>     
> >
> > They have a single fileserver which store all the data and you can
> > simple log into any of the diskless servers and find the same
> > environment. These servers just provide CPU power and the filesystem
> > server provides the storage.
> >
> > You could load balance, do cloud computing (like Google App Engine or so),
> > and you don't have a single point of failure, so if one server goes down
> > (except the fileserver), the members can simply use one of the others.
> > Moreover, new services don't have to be distributed to the server (like
> > http, smtp, imap or pop), because they operate on the same set of data,
> > this saves a lot of administration work and is more flexible. You could
> > simply PnP new servers.
> >   
> 
> It would be simple enough to let people opt into a variety of shared 
> filesystems.  I think most members would be glad for the chance to avoid 
> opting in, but we could still make it easy to access networked storage 
> from any machine.  Setting up the common daemons on a new machine should 
> be simple enough as to be an irrelevant cost, given the rate at which I 
> expect we'd add new member servers.  Maintaining them shouldn't be much 
> work, either.  It wasn't back when we had a single server.

If end up with this idea that members are bound to single server, then
we need a configuration management tool, so that we can easily replicate
configuration files to all the servers. Bcfg2 and Puppet look quite
promising, as an alternative domtool could be modified for this purpose.

Moreover, I suggest that domains are bound to single servers as well, so
that only http://www.hcoop.net/~user and user at hcoop.net have to be proxied
to the specific servers.

In addition to this we should provide some tools to migrate users to
other servers, if they wish (we could have an amd64 and x86 server).

If we do load balancing (via fcgi or so), we could put these domains on
the network filesystem. Opting in seems reasonable in this case.

Are we going to stick to Debian? Maintaining our own packages on Debian
seems to be quite uncomfortable to me.

I would volunteer to setup NFS, if decide to have some kind of network
filesystem, although I have never setup something bigger that the usual
home server.

Maybe we could make a poll about these decisions, because just deciding
on the future of HCoop with >5 people may cause us trouble afterwards.

Regards,
Matthias-Christian