[bind10-dev] BIND 10 OS support, was Re: BIND 10 Adventures on a Debian Super-Stable System

[Shane Kerr]

Anand,

Apologies for the long reply! I wanted to take this opportunity to get
some ideas out to the list.

On Mon, 2011-07-18 at 20:25 +0100, Anand Kumria wrote:
> On 18 July 2011 16:28, Shane Kerr <shane at isc.org> wrote:
> > I installed BIND 10 on box running the previous version of Debian stable
> > last week (Debian "lenny", version 5.0.8 to be precise). Debian is
> > always a bit stale, and running the previous version means "really old".
> > However, it will still be supported until 2012-04, and there are people
> 
> Does that mean that BIND 10 is expected to be released before then?

Yes, I expect BIND 10 to be production ready in 2011-01 for
authoritative servers, and in 2012-04 for recursive revolvers. 

> If not, and the criteria for BIND 10 is that it compiles on the
> previous version of Debian stable, why not target Debian 6.0 (lenny) -
> which is already fairly old. Note Debian 6.0 has two different kernels
> (Linux and kFreeBSD), will that impact BIND?

We actually don't have a specific set of distributions that we are
targeting. Rather we expect that the code should compile and run on a
wide variety of Unix and Unix-like systems. (We want to support Windows
too, but that is really a different topic.)

See below for more...

> > running various "enterprise Linux"; which as far as I can tell is a
> > codeword for "never change anything until I retire" (RHEL has a 7-year
> > 'regular life cycle' and another 3-year 'extended life cycle).
> >
> > Anyway, I don't know that we need to worry about installing on systems
> > from 2001, but I thought I would relate the issues I had getting BIND 10
> > installed on this system.
> 
> [snip]
> 
> Whilst I can, somewhat, understand the desire to see if things works
> since BIND 10 is being developed wouldn't it make more sense to spend
> time targeting existing systems and, once you get closer to feature
> completion, ask to see what your customers want to run BIND 10 on?

To be fair, Debian 5.0 is only about 2.5 years old.. not even 2 months
older than the BIND 10 project itself. :)

Anyway, the meaning of the word "customer" is a bit tricky here, as we
are making open source software.

First, the BIND 10 project has sponsors:

http://www.isc.org/bind10/sponsors

We've asked them what systems they use, as we hope that all of them will
some day use BIND 10.

ISC also has support customers (for BIND 9 and ISC DHCP), and we have
asked them to tell us what operating systems they use. FWIW, Solaris is
the most popular choice.

But there are also "customers" who are just random people running
computers everywhere in the world. We need to meet their needs too! When
deciding on operating system support, we basically need to support the
most popular systems. 

And finally, the developers on the project have their own personal
preferences. If an Amiga fan became an active developer, maybe we'd see
support for that. ;)

You can see our current build systems here:

http://git.bind10.isc.org/~tester/builder//builder.html

(Apologies for the weird URL, we just moved to a new server hardware.)

These are (in no particular order):

      * OS X
      * CentOS 5
      * FreeBSD 8
      * NetBSD 5
      * Solaris 10
      * Debian 5

We test with 32-bit/64-bit systems, Sparc/Intel CPUs, and gcc/Clang/Sun
Studio compilers, and with virtual/physical setups.


Even though we don't use Debian 6 in our build farms, I have run BIND 10
on it and it works fine. Some of our developers use Ubuntu, and it works
fine there too. I've run BIND 10 on Fedora and Arch Linux also, without
problem. At this point, we don't expect too many problems from different
server or desktop environments, although anything is possible... AIX and
HP-UX tend to be problematic from a portability point of view, new
compilers always spot different issues, and at some point pypy and
Jython will support Python 3 and break everything.

As I said earlier, Windows is a different discussion. We really want to
support it, but it cannot be a priority for us now since we have too
many other things to do. Our support for embedded systems is in a
similar state (although the longer we wait to support embedded systems,
the more powerful they become, and the easier the work).


Ultimately we want BIND 10 to be supported directly by distributors, the
same way that BIND 9 is. (I really want to be able to type "apt-get
install bind10-auth" instead of compiling it from source.) Each
distribution has its own way of doing things, and we can't know them
all, so we don't plan on being the package maintainers in most cases.
However, we would like to have a closer relationship with packagers for
BIND 10. This is probably going to be an area of active work for us in
2012, once we think the software is ready for general use.


Hopefully you're not more confused now than before, even though I know I
did not answer your question in a simple, direct way.

Cheers,

--
Shane