[Bioclusters] Debian cluster

Jonathan Barber bioclusters@bioinformatics.org
Fri, 19 Sep 2003 10:13:41 +0100

On Thu, Sep 18, 2003 at 10:28:57PM -0700, Matthew Laird wrote:
> The one comment I would like to offer is out of courtesy to the Debian 
> community I hope you will have a local mirror or the package archive or 
> subset you are using.
> On my personal machines I'm a Debian boy all the way just for the apt-get 
> reason.  But if you're planning to build a 50+ machine cluster using 
> Debian then run an apt-get upgrade on all of them when a patch comes 
> out... that could add noticable strain to the mirrors over time as well as 
> your institution's connection.
> I know on campus we have a Redhat mirror just for on campus use.  I also 
> have scripts to keep the subset I use on our cluster current and the 
> machines patched.
> This is definitely an avenue you might want to investgate for the courtesy 
> reason plus it's just faster to install from a local mirror.
> Another advantage I've learned about having a local mirror, especially 
> with Debian, is you can freeze the packages at any point you want so you 
> can always recreate the exact same configuration with no risk of package 
> upgrades conflicting with something.  That is one beef I had with Debian a 
> few years ago, mysterious package upgrades that sometimes broke other 
> packages, but I guess that's what you get when using the unstable release. 
> :)
> Anyhow, just wanted to pass along those lessons I've learned.  Good luck, 
> and let me know how Debian works out as the basis for a cluster.  I might 
> rebuild mine from RH9 to Debian if you have great successes. :)

Just to add some information, there is a dedicated drop in cache for
debian, apt-proxy, which will probably reduce the network load quite a
lot. Just alter your source's list to point at it and you're away.