[HCS-D]long-term planning

Matt Gline gline at fas.harvard.edu
Tue Dec 21 11:16:36 EST 2004


Okay, I've been thinking a bit about some long-term changes for the hcs
operational model, and I want to put them out here for discussion.
These ideas are founded only in my limited experience with other
existing systems, and so you should feel free to tear them apart with
your (collective) greater knowledge and rhetorical style.

1) I think HCS should move its home directory storage out of frog. My
gut intuition is that the expense of a netapp filer is, if we can cover
it with grants or other sources of revenue, worthwhile because they are
truly awesome devices. This seems like a good idea to me for a few
reasons. For one, moving storage out of frog will make it easier to recover
from issues like the ones we've been having. For one, whatever networked
storage alternative we come up with, even if it's just a separate BSD
server, will have snapshot directories, which are a valuable component
of any new backup strategy I can think of. And we can always mount up
/usr/home on a clean machine if hcs ever got toasted or broken into or
anything like that. 

2) I think HCS should move towards a kerberized environment. We've
talked about this before, there's some disagreement and it's fallen off
the radar but I think we should put it back on. AFS is really very cool,
and I think we should implement a general login environment using
kerberized LDAP (I pick LDAP largely on the basis of earlier
conversation this year, but if anyone has better proposals that aren't
NIS, I'd be happy to hear them)

3) I think HCS should move towards diversifying our services across
different servers, and making that transparent to our users. The fact
that lists and mysql are handled off hcs is nice, but a large part of
the benefit is wasted because when hcs goes down, mail sent to
listname at hcs doesn't get delivered. Once we have networked storage that
we can mount on several hosts, we can offload mail processing from frog
entirely and leave frog to be a web server. We might even consider
having groups log in to some machine that is not the web server, as fas
does... 

I'd like to know from people that know about these things first of all
whether these ideas seem reasonable as long term systems type plans go,
and second of all how they might fit in to the Debian-FreeBSD discussion
that's been going on in various places around the organization. 

My general sense is that deploying these things will make HCS more
robust, and will help on the backend towards accomplishing the goals
that Greg laid out at the meeting last night - namely, it will better
enable HCS to a) offer more services to our 'clients,' and b) enable us
to offer those services in a far more robust and reliable manner.

Comments appreciate, as I've made clear :)

--Matt


More information about the hcs-discuss mailing list