[tex-live] TeX Live on usb drive

[CTAN team (Jim Hefferon, TUG)]

On Thu, 2009-03-19 at 14:33 -0300, George N. White III wrote:
> I think some zip implementations will store files in "raw" form if the
> compressed version isn't smaller.    In any case, the real issue is to
> find a reliable way for users to get a complete copy, and for many of
> us, the usual mirroring by individual files is not straightforward.   Maybe
> and on-the-fly tar (e.g., without compression) would be better.

...

> It is a bit of a pain when different CTAN sites have different capabilities --
> you end up with user A advising user B to do something that works for
> the mirror site used by A but not the one used by B.

The situation is that we have zero control over our mirrors.  Indeed, we
are lucky to have them, and greatly appreciate their generosity.

A couple of years ago (?five ?eight) when our setup was different, I ran
a test to see which of our mirrors had both on-the-fly .zip and .tar.gz
working.  Out of the hundred or so that we had then, there were none
(except the three core sites) that had both working correctly.  (Just
for those who are interested in such things, the most common error was
to capture the stderr output of tar into the stream, so that when you
un-gzed you got a file whose first couple of bytes said something like
"Tar completed", followed by a correct .tar.  But when users did a tar
-xvzf they got "corrupt archive file".)  Since then, I notice that a few
mirrors have implemented them both correctly.

My point is not that mirror maintainers are doing it all wrong-- they
are not.  My point is that it is up to CTAN to make it work, not to this
group of volunteers who we can ask no more of.  Hence the .zip files
above many of the packages on CTAN (which, by the way, has caused Rainer
an awful lot of work).  

> The other issue is more intelligent redirection to steer users away from
> sites that have higher usage costs.   I think most people do try to follow
> the advice to use non-root sites, but it does make sense for root
> sites to run only the features used by non-root mirrors.

At this point we are actively trying to ship requests out to mirrors by
using the http://mirror.ctan.org/tex-archive/macros/latex/contrib -style
addresses.

In brief, on-the-fly is dead.

> It might be interesting to look at the ratio between the bytes served to
> CTAN non-root mirrors and the bytes served to users.

I don't believe that is possible.

Jim