Re: Debian has failed us
On Thu, 2007-01-11 at 11:40 -0600, Mathew Binkley wrote:
> Greetings. I am the senior system administrator at Vanderbilt
> University's supercomputing center. We operate a 1500 processor cluster
> for researchers at Vanderbilt.
> Our current cluster is divided between 840 Intel/AMD x86 processors, and
> 672 IBM PowerPC 970FX processors. To date, we have required different
> operating systems on each architecture because of poor OS support for
> the PowerPC's. SuSE has been our only option so far.
How so ? I'm routinely running debian/ppc on all sort of PowerPC
machines here, including js20's, 21's POWER5's etc...
You might need a bit of hacking to get the initial install though, and I
agree that should be improved in debian (though do they have access to
hardware for that ?)
Also, there are other options, like Fedora/RHEL, YDL, ...
> Etch supports PowerPC's running in 64-bit mode
Running in full 64 bits mode isn't necessarily the best idea on PowerPC
performance wise. It's generally recommended to run a 32 bits distro
with only the minimum set of 64 bits libraries so that applications that
have to be 64 bits can run. It's very different from AMD64 where 64 bits
brings all sorts of improvement starting with much more registers. On
PowerPC, we have many registers in the first place, thus 64 bits mode is
almost purely overhead unless you application has good use of 64 bits
pointers or arithmetic.
> so I was eager to try it
> on our cluster several months ago. I uncovered a small bug in the
> installer (the AMD 74xx driver was not compiled in the debian-installer
> kernel) which prevented me from installing Debian, and reported it to
> the PowerPC list in September 2006. I was promised that the driver
> would be included shortly.
Ah yes, I remember that.
> Unfortunately, some sort of personal issues between developers in the
> PowerPC group and the debian-installer group has prevented this minor
> fix from occuring. I do not understand what that problem *is*, only
> that personal problems are getting in the way of real-world results.
Yes, I've seen various mails/rants etc... and haven't wanted to take
part of it neither in the past, but it looks like this has gone too far,
and it's more and more looking like even perfectly good bug fixes that
are needed for most users are being rejected on the sole basis of the
person they originate from. Either that or somebody has an agenda of
fucking up powerpc support in debian...
> I had hoped to be able to install Debian on our cluster during our
> refresh this spring in order to have a consistent OS across all
> architectures. Since I have not been able to get this minor bug
> resolved in three months, we will be installing Centos 5 instead.
Can't you get a custom build of the installer ? I mean, I agree the
situation sucks, but it's not like you can't find somebody willing to
provide you with a custom build if you ask around ...
> I have personally been using Debian since slink, and I would like to
> believe that Debian can sort these issues out before it becomes a
> permanent handicap to the organization.
> Finally, I would like to extend thanks to Sven Luthor for tracking down
> the bug and trying to get me some assistance.
Luther, not Luthor, he's not that evil yet :-)