Hey all,
I'm new to the list, but I've been working with
ij on buildd.net. He methoned to me in passing
how Debian m68k was having trouble building
unstable due to the load, so I grabbed aranym,
and setup a buildd with distcc and a
cross-compiler. It's running stable, and
I've managed to build a few test packages with
sbuild (the machine itself is also listed on
buildd.net, as diablos).
Here are the specs
* Host Machine
MacBook Pro (hostname: siren)
2.16GHz Dual Core x86
Running Ubuntu Gutsy
2048MB of RAM
100GB of space
* Aranym Setup
Hostname: diablos
1024MB of RAM
1GB Boot Parition
40GB buildd parition
/proc/cpuinfo has been inconsistant on the
speed of aranym
The machine has a dynamic IP address, I have a
dyndns pointed at the machine so its always
accessable. However, for mail, and other services
requiring a static IP, I have the domain name
nemesisnetworks.com, and that has a mailserver on
it; mail is delivered to both siren and diablos
via fetchmail in multidrop mode; both can send
using nemesisnetworks.com as a smarthost.
Obviously enough, it isn't possible to directly
access diablos (unless I did some massive hacked
of my iptable scripts ;-)); you need to SSH into
siren, and then you can rlogin (or SSH if you
like waiting) into diablos.
If this setup is acceptable, I can also setup two
more machines in a few more weeks into the same
configuration.
Anway, a couple of notes and obvervations I made
during my setup of the buildd and such
1. With distcc, the compile part of building
packages is very quick (compared to just plain
complination), which does help cut down build
times. The problem though comes with
documetnation (I don't think there is much that
can be done about this unless someone wants to
write a distcc-like script for texinfo and
friends). It took about two hours to build the
documentation of zsh (about half an hour for each
format it took). There has got to be a
way we can reduce the time it takes.
2. dpkg-shlibdeps and sbuild take forever. I
suspose part of the reason is that they are both
perl scripts and they do run through the entire
APT package directory, but at least in
the case of shlibdeps, isn't there some way this
can be cached, or offloaded to another machine (I
suspose it is possible we could have a distcc
like setup here, but unless we had some
lightening fast m68ks, the only choice
for acceptable speed would be a machine of
another archiecture, and I don't think that's
fessible just because of the offhand chance that
a package might use different shlibs on different
archiectures).
3. Aranym can't be run without a GUI. I'm going
to dive into the guts of it sometime this week,
and remove this little limitation.
Anyway, I'm floating around #debian-68k, and I'm
available on jabber ([EMAIL PROTECTED]/Home)
if you want to reach me. I'm also available by
email.
Michael Casadevall
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
