Greg Hudson’s MIT blog

I’m currently putting minimal work into Athena, but have been thinking about it some. The last I wrote about Athena is here but a number of things have changed.

First, Linux survey information (not yet published) indicates that people want us to be supporting Ubuntu or Debian. Ubuntu’s most recent release supposedly does a better job of integrating the proprietary components you need for a functional desktop–proprietary video drivers, Flash, and so on. I’m still nervous about the practicality of their x86_64 support model (totally separate platform, can’t use 32-bit binary plugins to programs like Firefox), and I don’t know what an automated upgrade from RHEL 4 to Ubuntu would look like, but we’re currently leaning towards Ubuntu as the base platform for Athena 10.

I don’t know when Athena 10 is likely to happen. It won’t be this summer. This summer we are probably looking at a patch release with Firefox 2.0, proprietary video drivers on cluster machines for better graphics performance, and maybe the VMWare player if we can get permission. We’ve started getting complaints that Athena 9.4′s GNOME installation has bitrotted too far to build modern GNOME programs, and that’s only going to get worse over the coming year. So summer 2008 looks fairly likely.

In the past, we’ve tried to branch out to keep Athena relevant. I’d like to stop doing that, and just focus on making (with limited resources) a good system installation for cluster machines and quickstations, which is as close as possible to the underlying Linux distribution. I don’t think Athena can offer much to single-user desktops or laptops today, although I expect there to continue to be privately-administered Athena machines as long as we have clusters and quickstations.

So: an Athena machine is expected to be always on the network–ideally a nice, fast network. An Athena machine uses the MIT Kerberos namespace for logins and the AFS file store for user homedirs. (Some day MIT may transcend AFS, but probably not before it transcends Athena, unless a better distributed filesystem comes along. It might be possible to have clusters without shared homedirs, but I think it would be flaky and a lot of work.) Athena machines should generally take care of themselves, though a minority will be privately owned by administers who want to do their own tampering in addition to ours.

Lastly, Athena is not a concept or a set of services.  (My supervisor still talks about “Athena broadly defined,” but I’ll try to convince her to give that up.)  It’s a software installation that we’re supporting because, for the moment, MIT still seems to want to have quickstations and clusters.  It uses a set of services like AFS, MIT email, etc., but those are separate service offerings and not themselves part of Athena.