Re: RFC

[Jeffrey Watts <watts@jayhawks.net>]

On Wed, 15 Mar 2000, Robert W. Current wrote:

> I think you should read more of the comments from actual system
> administrators on these issues.

I'll bite on this one.  I _actually_ can discuss something that I know a
good deal about.  :-)

I admin six Origin2000 clusters and three Sun Enterprise clusters, all of
which run in-house telecommunications software.  I run a fairly complete
distribution of IRIX on the Origins, and a fairly complete setup of
Solaris on the Sun boxes.  As I said before, the apps are home-grown,
specialized telecom apps, none of which use X.

Is X installed on these servers?  Absolutely.  Why?  Most of my data
analysis and visualization tools (perfmon, Performance Co-Pilot) use X.
The GUI RAID tools are nifty and simplify some really complex tasks, as
well as making it much easier to visualize what you are doing.

My point is that even though my systems don't have a graphics console, and
don't have an X server binary installed, they do need the X libraries.

> Brilliant coded you may be, I have no doubt in my mind that most
> system admins would prefer to keep their "OS Updates" and "Software
> Updates" a lot more separated than most Linux distributions now
> actually do.

As far as separating out the updates goes, IRIX has an intelligent
installation tool and a nice "overlay" system in place.  It's far easier
to install the overlay as a whole and test it thoroughly in our integrated
test environment than it is to install individual patches, and figure out
the conflict manually.  I _prefer_ the dumb upgrade on the IRIX boxes over
applying individual patches to the Solaris boxes, because in general they
work better and have less problems.

Most RPM-based systems work similarly.  Given that a real sysadmin has a
test environment, it really doesn't pose a huge problem to do a "OS
upgrade".  If there are problems with certain packages, one can upgrade
from the command line only those packages that he/she wishes to upgrade.

I upgraded my home server by hand, and though it took a little longer than
just letting Red Hat's upgrade disk do it, it didn't take too much longer.
I honestly don't think that a professional sysadmin would really care --
most of us are going to do our own testing beforehand, and whether you are
upgrading five or twenty servers, you are going to be automating the
procedure, so the time required isn't proportional.

> Nope.  Either tell me that your going to close the spec, or make it
> logical why as a system admin I should accept installing X as base on
> 20 different servers for no good reason other than Donnie said "Deal."

Sysadmins are lazy by default.  A single Xlibs package isn't a big deal,
and it's easier to have it installed than to not.

"Install 'em all, and let Integrated Test figure it out"
-- Jeffrey Watts

Hope this makes sense, I'm a bit low on sleep.  :-)

J.

o-----------------------------------o
| Jeffrey Watts                     |
| watts@jayhawks.net         o-------------------------------------------o
| Systems Programmer         | "At the beginning of the week, we sealed  |
| Network Systems Management |  ten BSD programmers into a computer room |
| Sprint Communications      |  with a single distribution of BSD Unix.  |
o----------------------------|  Upon opening the room after seven days,  |
                             |  we found all ten programmers dead,       |
                             |  clutching each others' throats, and      |
                             |  thirteen new flavors of BSD."            |
                             |  -- Unknown                               |
                             o-------------------------------------------o