[1402] in SIPB-AFS-requests
Re: Reorganizing AFS machine configuration directories
daemon@ATHENA.MIT.EDU (mhpower@MIT.EDU)
Sun Jun 5 01:35:21 1994
From: mhpower@MIT.EDU
To: ghudson@MIT.EDU
Cc: star-maintainers@MIT.EDU, webmaster@MIT.EDU
In-Reply-To: "[28975] in Usenet_Meeting",
"[0834] in Charon_Maintainers_Archive",
"[1673] in RTFM_Maintainers_Archive",
"[2180] in Webmaster",
"[1400] in SIPB-AFS-requests"
Date: Sun, 05 Jun 94 01:34:49 EDT
>At any rate, I'm writing here because, as Bert and other have pointed
>out, the way we mirror machine configurations in AFS is a complete
>mess:
Greg,
Creating a new mailing list that, among other things, feeds a
public discuss meeting, and then sending mail to it characterizing
the work of other SIPB members as "a complete mess" seems to me
exceedingly unhelpful. (I'm trying to be consistent with the
gratuitous spewing from the related "Re: screwed up acls" thread.)
I agree that, with the new server arrangement, we'll no longer want to
keep files in something called project.foo-server. That was originally
set up because we had two server machines that happened to be running
the same operating system (which made for a lot of similar/identical
configuration files), and, probably more importantly, happened to be
maintained by, more or less, the same group of people. I don't think
this will be true much longer. The only similar situation might be
having the same people maintain both rtfm and the discuss/nfs server,
but those will have different operating systems, the configuration
files probably won't overlap much, and so there's not much advantage
(e.g., even in saving disk space) to having a "common" directory.
Because of this, I definitely agree that your creating a project.rftm
volume and moving files there has been a good idea.
However, I think creating project.servers is not a good idea. The
different server machines are, to a much larger extent than before,
maintained by different groups of people, and it seems suboptimal for
everyone to be sharing the same afs volume, since it's too risky to
have to depend on global cooperation/communication about what files
belong where, who should have access to change them, etc.
I'd prefer having separate afs volumes for each server machine
(probably excluding the news servers, though), have each volume owned
by whatever group maintains the machine, give it its own disk quota
(so it's not affected by disk usage of other machine's maintainers),
and arrange the files in whatever way makes sense to the people who
maintain that particular machine. I think this will vary (as it does
now) and that's entirely appropriate. Maintaining files in afs with
RCS is probably the safest approach, but that takes more time. Not
everyone will want the same tradeoff. It might depend on whether we
have 2 people maintaining the machine, or 12. It might depend on
whether the people have 8 years of experience, or are prospectives who
showed up last week. It might depend on whether the maintainers of
that machine discuss the setup in person fairly often, or whether they
sometimes remember to send mail if they replace the kernel.
Also (although this isn't a major point), I think the distinction
between "server" and "service" volumes will end up being vague and
misunderstood, and people will end up randomly trying both places.
But if you want to try keep them separate, that would be ok.
Matt
