\title{Moving the UK CTAN}
\author[Martyn Johnson and Robin Fairbairns]{Martyn Johnson (\texttt{maj@cl.cam.ac.uk})\\
        and\\
        Robin Fairbairns (\texttt{rf@cl.cam.ac.uk})}
\begin{Article}
\section{The background (RF)}

The UK node of the Comprehensive \TeX{} Archive Network (\CTAN) has a
long and honourable history, which starts long ago in the recognition
(by Peter Abbott at Aston~--- see \cite{abbott:exeter}) of the need to
provide an archive of \TeX-related material within the UK.  At the
time that the Aston archive was created, \TeX-related material was
mostly made available \emph{ad hoc} by its originators~--- there was
no site with ambitions to provide a \emph{complete} set of systems,
macros, and so on.  Furthermore, access from within the UK to overseas
material was less than straightforward (access to ftp, using an
account at UCL, was severely restricted).

The Aston archive was originally a VMS-based facility, offering
connection via the Janet coloured-book protocols to machines that were
part of Aston's centrally-provided service.  Later, the archive group
were given a second-hand VAX, and later still a parallel version of
the archive was established on a SparcStation that sat on Peter
Abbott's desk.  This machine (with the net name |ftp.tex.ac.uk|)
eventually became part of \CTAN, offering access via anonymous
ftp to all and sundry throughout the world.

At the beginning of this year, Peter Abbott told your committee that
he would be retiring (early) at the end of July; the implication was
that it would be unlikely that we could count on Aston's willingness
to offer a home to the archive beyond that date.

The committee discussed whether it was reasonable even to consider
maintaining a CTAN node in the UK (covering the world with two sites
in Europe and one in the USA is hardly a convincing approach); and if
the node was to be retained, the implications of so doing.  After much
soul-searching, we decided that it wasn't reasonable to spend \UKTUG{}
funds on a new home for the archive; we concluded that our membership
is such a small proportion of the \TeX{} community within the
UK\footnote{In contrast to the situation, say, in Germany, where
\DANTE{} owns the archive machine.}, that group funds really
\emph{shouldn't} be expected to cough up for support of the community
at large.

In parallel with these discussions, the committee investigated
alternative new sites for the archive; candidates were the
Universities of Warwick, Sussex and Cambridge, the National
Typesetting service (at Oxford) and the Hensa\footnote{Higher
Education National Software Archive.}.  While the typesetting archive
looked promising at first, they eventually
suggested that we should approach Hensa (which had already been on our
list of possible candidates).  Hensa in fact maintains two archives,
one for micros and one for Unix; since \CTAN{} is a cross-architecture
service, offering support for micros, Unix machines and others, it was
obvious that Hensa couldn't maintain a \CTAN{} node.  Sebastian Rahtz,
who's the only volunteer effort that the UK \CTAN{} node has, was
unwilling to maintain an archive that wasn't \CTAN, so we decided to
look elsewhere.

Warwick (in the person of Malcolm Clark~--- he of the
\emph{Gleanings}) concluded that they would need to be provided with
some extra disk space before they could offer to host the service.
Sussex would very much like to host the service, but were unwilling to
do so until their connection to SuperJanet went live.  Cambridge
expressed an early interest, whereafter very little (that was visible
to the committee) happened for some time while Martyn Johnson
established what was necessary and acquired agreement from the head of
the department and from the rest of the Computer Laboratory's systems
group.

\section{The Archive Operational Requirement}

Towards the end of June Sebastian Rahtz wrote an `operational
requirement', which described the facilities provided by the Aston
archive, and explained which were essential and which might, at a
pinch, be dropped.

The complexity of it all was quite a surprise.  The primary service is
an ftp server, but it needs to be able to perform many automatic
transformations, such as packing a directory into a |zip| image.
Gopher and WWW services were also provided at Aston.  Behind the
scenes, there was a mechanism which kept the main \CTAN{} nodes in
step with each other whilst maintaining a peer relationship between
them. The archive maintainers needed remote access to the machine and
the ability to manipulate the archive.  Finally, there were several
well-publicised mail addresses at |ftp.tex.ac.uk|.

\section{Meeting the Operational Requirement (MAJ)}

Whilst it was clear that we had the resources available to run the
archive, there were a number of awkward decisions to be made. The main
question was ``which machine?''; we certainly didn't have a machine
available to dedicate to the task.  My initial assumption was that I
would put the service on one of our main fileserver machines, a
Digital Alpha/AXP 3000/500S, since that machine had plenty of disc
space available.  But I was worried about security~--- we don't even
allow our own users to log in to our fileservers.  Then I realised
that one of our most public machines (another Alpha) had a 1Gbyte disc
which could be made available, and the decision was made.  A welcome
side effect of this choice was that it made it easier for me to
provide external access to the archive through NFS without
compromising the security of the internal services.

Transferring the data from Aston to Cambridge was the least of the
problems.  Not very long ago, moving such a large amount of data over
the network would have been unthinkable.  I just did it, using a
standard ``mirror'' script running in the background.  It took about a
day to pull the archive across, and a few minutes each day thereafter
to keep it up to date.

It was obvious that the standard ftp daemon supplied with OSF/1
wasn't up to the job of running the archive. The only sensible choice
was \texttt{wu-ftpd}, as used at the other \CTAN{} nodes. This
compiled easily enough for the Alpha, and initially appeared to work.
However, none of the document conversion scripts would function at
all, and the bug\footnote{Overwriting the daemon's own argument and
  environment strings for cosmetic reasons, so that subprocesses start
  up with a corrupt environment which the OSF/1 shell takes umbrage
  at.} was only found after a great deal of detective work. Once found,
the problem was easy to fix, but even then, one of the document
conversion scripts still didn't work\footnote{But it didn't work at
  Aston either, and has been mended.}.

After those and other bugs\footnote{Notably the bug that made
  \texttt{zoo}'s `portability library' not port to 64-bit machines.}
had been mended, there were still the Gopher and WWW services to
consider.  We had not previously run a Gopher server, and I was not
very keen on doing it, but Sebastian assured me that the service was
well used and would be missed if we didn't provide it.  In the end the
Gopher service turned out to be straightforward to make available,
though we decided to not to attempt to set up WAIS indexing initially.

WWW was even easier, since we already ran a suitable server on another
machine, and we merely had to copy the data into it.  Or did we?  All
of the Aston archive's services were published as being available from
the machine |ftp.tex.ac.uk|, and this could obviously not be made an
alias for two different machines.  I really did not want to run a
second WWW server, so we decided that the world would just have to
change.  Each service offered by the archive now has its own name:
|ftp.tex.ac.uk|, |gopher.tex.ac.uk|, |nfs.tex.ac.uk| and
|www.tex.ac.uk|.  While we were at it, we thought that a better name
for the mail domain would be simply |tex.ac.uk|, though we intend to
continue to accept mail using the old addresses for some time.

Of course we cannot force people to use the correct name. In fact only
the mail and WWW services are provided on machines other than
|ftp.tex.ac.uk|.  Mail is not a problem since there are
well-established mechanisms for redirecting it to a site hub.  In order
that the WWW service should not seem to have disappeared to people
following old links, I wrote a tombstone service.  No
matter what you ask for, it always returns a fixed page, which
explains that the server has moved, and offers a link to the new
place.

The mail facilities were provided by configuring a logically separate
mail system within our mail hub, which runs the PP mailer. We did not
want to do anything which would make a further move of the archive
more difficult, so we were very keen to ensure that users could not
confuse the mail domains |tex.ac.uk| and |cl.cam.ac.uk|.  We cannot
prevent them being confused, but at least we can ensure that they will
be told that they are confused by having incorrectly addressed mail
returned to them.

It was always part of the agreement that the remote management of the
archive by Sebastian Rahtz, David Osborne and others should continue.
It is easy enough to give those users login access to the archive
machine, but they needed to be able to manipulate the archive without
needing ``|root|'' privilege.  The archive is owned by a pseudo-user
``|ctan|'', and the archivists do their work using a local command
which allows them to pretend to be |ctan|, provided that they can
quote the password.  Much of the maintenance of the archive is handled
automatically, by passing mail around and running periodic jobs.  All
we had to do was arrange to deliver the mail~--- Sebastian did the rest!

All of a sudden, it seemed that we were ready to go live.  We had
already taken over control of the |tex.ac.uk| domain of the DNS
(a saga in itself) and simply had to flip over the addresses and watch
the traffic come in.  And that is exactly what happened.

Three days before he retired, Peter Abbott sent a message to the
committee mailing list saying that Aston did indeed want to reallocate
the Archive machine from the day that Peter left; the move was `just
in time'!

\section{The first six weeks (MAJ)}

We received remarkably little mail concerning the changeover, so it
must be considered a success.  A handful of people had bound to the
old address and needed to be told the new one, but it seems that the
vast majority of clients simply did not notice any change.  One person
complained that the Gopher service was not quite what it was, and this
is indeed an area which needs some more work.  So far, after nearly 6
weeks of operation, the new archive has shipped some 11Gbytes of stuff
via ftp.  Equally encouragingly, there have been no complaints from
local users of the machine, for whom it is a compute server, that the
archive activities are having any impact on their work.  In fact the
Alpha is a very fast machine, and provided that there is enough
memory, a few ftp sessions are hardly noticeable.

By a curious twist of fate, the service offered in the first few weeks
has not been as good as we might have hoped.  One of the reasons we
chose to use the Alpha machine we did was that in practice its
hardware and software have been highly reliable.  Unfortunately there
have been three serious problems since we transferred the archive.

The first was a spectacular thunderstorm in the Cambridge area which
caused widespread disruption to our equipment.  The server itself
recovered with little difficulty, but the network problems were more
serious and our whole department was cut off from the outside world
for some time.

The next weekend, the archive machine itself failed.  The front
panel lights were dim and flickery, but this `obvious' power supply fault
was actually being provoked by a faulty system disc, so a replacement
had to be sent for, and the system had then to be restored from backup
tapes.  This meant that the service was down for most of the weekend
and the following Monday.

The third failure again started during a weekend.  The machine simply
froze, and when rebooted, did exactly the same thing again.  The
problem was tracked down to the NFS server, which was being fed
``poison packets'' by a machine in Norway.  The machine's
administrator didn't respond to mail, but back-door intervention on
the archive machine had the required effect.  We're told that the bug
in OSF/1 which made it crash on receiving the bad packet is mended in
the next release.

Fortunately the user community has been very tolerant of these early
problems, and we very much hope that the next few weeks will be better.

\section{Conclusion (RF)}

The archive has grown from humble beginnings as a side service on a
VAX/VMS machine using protocols only (significantly) available within
the UK, to become part of an internationally-coordinated group of
archive sites offering services to anyone who chooses to use it in the
world at large.  As a community, we have many reasons to thank Peter
Abbott for his foresight, and Aston University for hosting the archive
for many years.

For the time being, Cambridge University has taken over the torch; let
us hope that the \CTAN{} network can continue to serve as a beacon
leading progress in the use of \TeX{} world-wide.

\begin{thebibliography}{\bf{Abbott}90}

\bibitem[Abbott 1990]{abbott:exeter}
Peter Abbott.
\newblock {\em {UK\TeX} and the Aston Archive}.
\newblock In Malcolm Clark, editor, {\em {\TeX} Applications, Uses, Methods},
  pages 109--114. Ellis Horwood, 1990.
\newblock Proceedings of {\TeX}eter (1988).

\end{thebibliography}

\end{Article}