[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Are mails sent to xxxx <at> buildd.debian.org sent to /dev/null ?

On Thu, Dec 16, 2004 at 04:59:15PM -0600, Dirk Eddelbuettel wrote:
> On Thu, Dec 16, 2004 at 11:34:55PM +0100, Ingo Juergensmann wrote:
> > Although the problem is well known and the solution is obvious, nobody seems
> > to have the guts to make a change (or even to speak about it).  
> Let's have a discussion about reducing our number of architectures.

The number of archs introduces some additional work, true, but it's usually
no showstopper. 

> Attempting to support several thousand binary packages on a dozen
> architectures across three release flavours imposes a large cost (see e.g.
> Joey Hess' recent blogging about the increased sysadmin work he has to do so
> that he can test d-installer).

Of course, when you do such things as a new installer, you have additional
work in porting it. Another choice would have been to use old boot-floppies
on those archs that are not that well supported by d-i. But it was a
decision made by someone that all archs should use the new d-i. Complaining
afterwards that your own decision puts some more work on you is somewhat
strange, eh?

> So we know the costs. Can we quantify the benefits?  

> In http://blog.bofh.it/id_66, Marco showed these numbers of downloads at the
> Italian site:
> architecture files downloaded
> i386	     1285422
> all	     504789
> powerpc	     17754
> ia64	     10111
> sparc	     3336
> arm	     850
> alpha	     507
> hppa	     204
> mipsel	     91
> m68k	     15
> mips	     7
> s390	     4
> total	     1823090
> AFAIK this data has not been refuted. 

So, let's begin with throwing out s390 and mips... 
> Clearly, for informed decision we'd need more data, from more hosts and over
> longer periods.  

How about number of installed packages? Another funny statistic:

5280 : mipsel
5428 : mips
5482 : alpha
5514 : hppa
5538 : arm
5546 : s390
5565 : powerpc
5578 : m68k
5593 : sparc
5601 : ia64
5680 : amd64
5973 : i386

We could argue that the low numbers of the users in popcon might be caused by the low
number of available packages. So, throw out mipsel, mips, alpha, hppa, arm,
s390 and powerpc.  
> But it would be nice if we based this discussion on /measurable usage/ of
> Debian to complement the costs with actual benefits -- as opposed to
> hypotheticals (such as compiling thousands of packages for arches without
> actual users).

I know you and Marco are great haters of m68k & co, just because you
sometimes need to wait one or two days longer for your package. But may I
remind you on the qt-x11-free story from the begin of this year on mipsel?
That was a great blocker for about >70 other packages caused by a ignorant
buildd admin. 

> As it stands, 4 downloads for s390 appear somewhat disproportionate to
> 1,285,422 for i386.

As always for statistics, they mean nothing without a proper interpretation
- and how you interprate the data is totally up to you. For me, it only
shows that there are not many s390s in Italy or they don't use the italian
mirror for whatever reasons or they don't update their systems that often
because they run stable instead of unstable as many of the i386 users do. 

How about solving the problems that exists and where the solution doesn't
punish our users (or maintainers)?

Ciao...              // 
      Ingo         \X/

Reply to: