Roadmap for UbuntuServer

Tue Feb 21 22:07:29 UTC 2006

Since I must pick just one message to reply to...

On 2/20/06, Joao Inacio <jcinacio at gmail.com> wrote:

> > As you can see everything isn't black and white.
> >
>
> that's exactly my point.
> though i believe that if there is a good-enough selection of server
> software included in main, creating some meta-packages for installing
> and pre-configuring these packages would be a good starting point.

Exactly so.

It is not easy to pick the "best" package for a job, no. But this is
true on the desktop as well - nonetheless, the Ubuntu team did it.
Certainly some people would prefer AbiWord and Gnumeric to Openoffice;
some would prefer Opera or Seamonkey to Firefox; some prefer KDE to
GNOME.

But to make something that a non-expert can use, it's necessary to
simplify the huge range of choices of open-source software down to one
"best" application - one program that does the job, offers the core
20% of the functionality that 80% of the people need.

http://en.wikipedia.org/wiki/Pareto_principle

OpenOffice isn't perfect, but it works, it does the job, it can read
and write Microsoft files - so in OpenOffice went.

GAIM may not be the best instant messenger, but it talks to all the
main services and it is easy to use so in it went.

And so on.

We (if I may say that) can do the same thing.

We can say, "look, if people want a LAMP server, then they will need
Apache, MySQL, PHP, and Perl", and include them, and put a management
screen on the front.

We can say, "for a file server, they will need Samba, SWAT, netatalk
and nfs-utils", and include them, and put an admin screen on the
front.

For a print server, CUPS and Ghostscript and admin.

The only one where the choices would be difficult would be for a mail
server, but I do not this this is insuperable. Sendmail is too hard,
Qmail is controversial due to its development methods, and so on.

There is nothing to stop a range of packages being included in the
repository, for experts to install whatever they wish, but that alone
is not enough.

Not everyone who wants a network server is an expert. If Ubuntu itself
had just offered a set of refined Debian packages and a base install
and left people to build their own desktop, it would not have got
anywhere. Few would have bothered because very few have the skills.

If we want Ubuntu Server to have the same kind of success as the
desktop version has been, we need to make it a lot simpler for people.

Here is a true story, by way of illustration.

Recently, I have built myself an Ubuntu fileserver for my home
network. It took me many weeks of effort, consulting more
knowledgeable friends, fiddling with configurations, reinstalling and
reinstalling again and reinstalling again. Debian 3.1 with the 2.4
kernel couldn't see all my RAID drives; Debian 3.1 with 2.6 could, but
kernel panicked when it loaded the network driver; ClarkConnect
wouldn't even install; SME Server doesn't support RAID5 directly;
finally, Ubuntu did it, but then I had to work out what packages to
add for SMB, then how to mount my RAID, then what permissions to give
the mount point and the RAID filesystem itself, and how to configure
my user accounts so that they could write to the share.

This took me more than a month of work in my spare time. I still don't
have classic MacOS support or NFS support because I don't know how to
do that. What I need most is Windows support - Mac OS X and Linux can
connect using that, even if it isn't very efficient.

And I have been using Linux for 9 years now and Unix for nearly 20
years. I have put in dozens of Unix and Linux machines in live
commercial environments. Linux has been my main desktop OS for more
than 2 years now.

These kind of things are /hard to do/ on a server if you do not
already have the expert knowledge.

At the moment, Ubuntu Server does not install a server. It installs a
toolkit and leaves you with access to a kit of parts, and the admin
must build the server themselves.

As such, there is little reason for anyone to use it; if they have the
knowledge to do that, why not use Debian itself? It at least will
pre-install a set of components for a fileserver or a webserver or
whatever for you! It just leaves you to /configure/ it - and my home
Debian webserver still is not live after more than 6 months, because
I've not yet worked out how to make it possible for me to upload
content to it from another machine.

> experts are still free to choose their prefered applications and
> manage everything "by hand", but any user wanting to set up a simple
> web/mail/whatever server could start by just installing a
> meta-package.

That is a start, but even if we do that, we are still not offering
anything more than raw Debian offers.

> also, i think software like webmin[1] is a very valuable tool to have.

I agree with that. It is not perfect, but it is a lot more helpful for
a novice administrator than a bash prompt. I run Webmin on all my
servers. Because the default Debian installation leaves Webmin
unusable, with no actual modules installed, I install Debian's webmin
package to fulfil any dependencies, then I remove it again, install
Links and go and fetch the  latest Webmin tarball from webmin.com and
install that myself.

But if people don't like Webmin, let's hear about alternatives! Or
would it be preferable to design and build a web-based management
system from scratch?

I really think this is worth doing. As far as I know, nobody in the
Linux world is doing anything quite like this as a free product, and
so if it were done right, it could be hugely successful. It will be
good for Ubuntu and good for Linux and open source software in
general.

--
Liam Proven · http://livejournal.com/userinfo.bml?user=lproven
Tel: +44 20-8685-0498 · Mob: +44 7939-087884 · Fax: +44 870 1271528
AOL/AIM/iChat: liamproven at aol.com · MSN/Messenger: lproven at hotmail.com
Yahoo: liamproven at yahoo.co.uk · Skype: liamproven · ICQ: 73187508
Gmail/Google Talk/Orkut: lproven at gmail.com