Cluster Poll Results (tangent into OS choices, Fedora and Debian)

Andrew M.A. Cater amacater at galactic.demon.co.uk
Sat Nov 1 05:05:07 EST 2003


On Fri, Oct 31, 2003 at 06:59:16PM -0500, Tod Hagan wrote:
> 
> If Fedora achieves 2-3 upgrades per year then it will be fairly
> different from Debian, which seems to be at 2-3 years per upgrade these
> days, (well almost).
I think it's averaged out at about 18 months overall for each major 
version release.  Point releases of security fixes come out more 
frequently. Debian 2.2 was there for about two years with about 7 point 
releases, the last being made days before 3.0.  3.0 has only had one 
point release - but security fixes and so on are updated quickly.
> 
> After a new release comes out Debian supports the old one for a period
> of time (12 months?) with security updates before pulling the plug.
> 
Given a two year release cycle, that means you may get three years of 
full support.  We don't kill things off on fixed dates, necessarily, and 
it's open to every package maintainer to build fixes for "old stable" 
for as long as he wishes.  One aim is to support upgrades from older 
systems easily: I'm fairly sure you can go from 2.1 - 2.2 - 3.0 - 
unstable with about four reboots - so thats about six or seven years of 
development in a couple of hours :)
> Debian can be upgraded in place as opposed to requiring a full
> resinstall; while this is great for desktops and servers, I'm not sure
> if this is important for a cluster.
Upgrades are relatively straightforward - unless you change kernels / 
a.out -> ELF / glibc major versions, you probably don't need to reboot.
>
> As a result of the extended release cycle Debian stable tends to lack
> support for the newest hardware (Opteron 64-bit, for example). This is
> why Knoppix, which is based on Debian, isn't derived from Debian stable,
> but rather from packages in the newer releases (testing, unstable and
> experimental). But the flip side is that the stable release, while
> dated, tends to work well as it's had a lot of testing.
Debian also works on 11 hardware architectures, with more coming along.
We've had 64 bit issues on Alpha, Sparc and Itanium for a while. The 32
bit distribution works well on Opteron but there is also 64 bit stuff 
working.  Testing is creeping asymptotically to release, as ever :)

> 
> Debian could probably use more recognition as a target platform by
> commercial software vendors but it incorporates a huge number of
> packages including many open source applications pertinent to science.
> Breadth in packaged applications is probably more important for
> workstations since clusters tend to use small numbers of apps very
> intensely.
>
There's a lot of stuff packaged by Debian people who want, for example,
genome sequencing / heavy maths and other "stuff" :) 
> 
> Wow, I guess I just slung some FUD at Fedora, but maintaining a 2-3
> releases per year rate probably requires a small core, putting the bulk
> of applications into the Extras category and thus increasing the chance
> of conflict. (Wasn't that the original recipe for DLL hell?) Debian has
> avoided this through a much larger core, which of course slows the
> release cycle.
> 
The key is tight dependency control and management. That's what has set
Debian apart from the distributions based on .rpm.  There's a heavy 
overhead for the maintainers but hopefully a lighter burden on users. 

[Full disclosure: I'm also amacater at debian.org :) ]

Andy
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list