On 1 Feb 2007, at 1:00 am, Charles Plessy wrote:
(Sorry for the noise, I reply on the list since Sanger's mail server thinks I am a spammer.
Does it?
I am very interested to hear that Sanger is using Debian on thousands ofmachines. Do not hesitate to tell us a bit more ondebian-med@lists.debian.org. Bioinformatics is part of our effort as weinclude it in "pre-clinical research", and we would love to see our priorities a bit more user-driven.
Well, we basically use Debian everywhere we can; we have a few machines running other Linux variants where support requires it (Oracle, mainly).
Our Top 500 compute cluster runs almost entirely Debian, this consists of 600 machines. Two are Tru64 Alphaservers, two are SGI Altix 350's running a horrible old version of Red Hat, and the remainder are all IBM blade servers running Debian. A total of 1462 CPUs.
All the Debian machines use the HP SFS cluster filesystem (HP's productized version of Lustre) for sharing data. Job scheduling is with Platform LSF, and configuration management is with cfengine2 (our own modification of this; the package currently in etch lacks some features we need).
Of the 700 or so desktop machines in the Institute, about 300 are running Debian, the rest Windows. There is a possibility we may move the desktop machines to Ubuntu, but that is not decided yet. Again, configuration management, in common with the cluster, is through cfengine2.
We maintain our own internal package repository for both home-grown packages and local modifications of standard debian packages, backports and whatnot.
We're quite keen to present something about all this at Debconf; I realise the deadline has passed, but hopefully they'll squeeze us in...
For the moment, I am finishing to improve support of multiple alignmentprograms, and will work on EMBOSS with others soon. http://wiki.debian.org/SequenceAlignment http://alioth.debian.org/projects/pkg-emboss/
Generally, we are not using Debian packaging techniques for bioinformatics software. The software requirements of the users move too quickly, so packaged versions are always out of date, and besides, different users often require different versions, so we just leave the maintenance of such software packages to the users themselves, and store them on a central BlueArc NFS server.
That's not to say that your packaging efforts are not valuable; I think they're extremely valuable, and will help small laboratories and the like build functional bioinformatics systems very quickly without requiring large amounts of assistance from dedicated support staff they probably can't afford.
Regards, Tim