[tex-live] SysAdmin Stuff (Was: Re: tl09 release status: coming up)

C.M. Connelly cmc at math.hmc.edu
Tue Oct 13 00:35:57 CEST 2009

<#secure method=pgpmime mode=sign> 

(Changing the subject line, as this message is veering pretty far
away from the original topic.)

"SK" == Siep Kroonenberg <siepo at cybercomm.nl>
"CMC" == C.M. Connelly <cmc at math.hmc.edu>
"RF" == Robin Fairbairns <Robin.Fairbairns at cl.cam.ac.uk>

    SK> E.g. if you sysadmin doesn't give you enough rights on
    SK> your computer.

    CMC> I can assure you that if one of my users installed TeX Live
    CMC> in their own account, I would be very, very unhappy with
    CMC> them, and I'm fairly sure I speak for most sysadmins.

    RF> depends what you mean by "their own account" (it's
    RF> ambiguous).

I meant in their home directory.  We actually trust people so much
that we don't have quotas, but we also don't have huge amounts of
space on the server so good citizenship is a must.

    RF> i wouldn't mind in the slightest if they chose to install
    RF> it on scratch space on their own computer ... except
    RF> insofar as they failed to install it and expected me to
    RF> pick up the pieces.

Right, which is why I'm not enthusiastic about letting people
install random things on machines I manage.  If they want
something, and it's sane, I give it to them.  If not, I explain
why and encourage them to try it on their own machines (or on
their laptops, which are less locked down).

As we're the math department, TeX is, of course, a must, and any
machine I build is going to have some nice TeX system on it.

    RF> it may be different in other departments, but i understood
    RF> that computer science people in the states typically have
    RF> more freedom still than those in the computer lab here.

Yes, our CS profs actually manage their own desktops, with
attendant hijinks.  My faculty don't, except for the tiny number
of Windows users, who are mostly not supported by me.  I'm pretty
sure that the CS folks only get to talk to their servers over
user-level connections, though (e.g., not NFS mounts).  If the
Windows machines were more tied into our servers, I would probably
not let them have administrative rights on the Windows machines,
either.  And if we were less dependent on NFS on the other
machines, I might be more open to giving them more control.

    RF> we expect these people to be sensible.  mostly they are,
    RF> and many who don't like swimming out of their depth do
    RF> come and ask, but the occasional one takes great advantage
    RF> of the rope we provide and does a good job of hanging
    RF> hirself.  we come along, then, and cut them down.  and ask
    RF> what they _thought_ they were doing.  (occasionally that
    RF> sort of thing provides input for the faq...)

Yup.  There's a balance point that you have to find that works for
your particular situation.  For me, the ``extra'' work involved in
making sure people get the software they need installed is a small
price to pay compared to the possible problems that I might see if
the machines were less controlled.

The key, I think, is that we're a small enough department that I
can be pretty responsive about installing software people need
when they ask for it.  In practice, that approach seems to be
working quite well for us.

    CMC> Definitely make it easy to use from a DVD or USB drive,
    CMC> but please *don't* make it easy to install in
    CMC> individuals' accounts.

    RF> damnably difficult -- you have to tread a path between
    RF> usability (in any sense) and discouraging some people.
    RF> personally, i would prefer the team to err on the side of
    RF> usability: i'm no superman, and i don't want to spend my
    RF> time wrestling with one of those stupid knots tied by so
    RF> many software package designers.

Agreed, which is why I'm generally a fan of letting the
distributions do the packaging rather than having it done
upstream.  The distro packagers know how their systems work, where
files are supposed to live, and how to get their packaging systems
to work.  As they're all different (I'm a Debian developer and
also package a lot of software for our CentOS machines), it's
better for upstream to concentrate on the core software and not
worry about all the packaging stuff.

Windows, of course, is always the problem child, as there don't
seem to be many (any?) third-party packagers in that space.


  Claire Connelly                              cmc at math.hmc.edu
  Systems Administrator                          (909) 621-8754
  Department of Mathematics                 Harvey Mudd College

More information about the tex-live mailing list