When I wrote my previous Redhat upgrade memo, my recommendation that we move to Redhat Enterprise over Fedora was based on these notions:
It's beginning to look to me like both of these notions may be false.
I think that the primary interest of the ZEUS, Muon, and Neutrino groups in the Nevis Linux cluster is that the standard physics tools are available: ROOT, Geant4, OpenOffice, and (for a while) CERNLIB. All of these tools either work or can be re-compiled for both Fedora and RHEN.
My impression of the DOE group is that its use of the cluster is the same as that of the other groups: if ROOT continues to work, then the version of Linux is irrelevant.
That leaves ATLAS. I've installed Fedora Core 1 on my desktop machine tanya. I've confirmed that the existing Geant4 libraries (compiled on RH9) run on Fedora. The GCC 3.3 compiler on Fedora is slightly less forgiving than GCC 3.2 about deviations from the C++ standard, but the differences were easy to fix. There are issues with running the ATLAS software framework, Athena, on Fedora systems, but these issues are the same as running on any system without an official CERN software distribution.
Another issue is the purchase of Redhat Enterprise. We'd need about $1000 of RHEN Academic licenses at Nevis; however, Redhat will not take a purchase order for less than $2000. The Columbia University AcIS Site License Group is looking into a university-wide RHEN license, but Redhat has not been anxious to pursue the deal (probably because, rightly or wrongly, Redhat perceives that their markets are large companies with banks of servers, not educational institutions).
Putting it all together:
I no longer think there's a need to rush to any decisions. The Fedora Legacy project will continue to supply security patches to Redhat 9 for another 1.5 years.
So my intention is to do what we've been doing up till now: New systems get the latest Fedora Linux distribution available, and so we adiabatically upgrade the systems. There may be exceptions; for example, Fedora Core 2 will use the Linux 2.6 kernel, which we may not want to install on systems that require specific drivers for, e.g., electronics testing. We only do "forced upgrades" (as I did last December, when I upgraded all the systems to RH9) if there will an end-of-life to the security support for a particular system.
There are disadvantages to this approach:
However, I think we can deal with these problems as they arise.