[Discuss] upgrade-itis

[Alan W. Irwin]

On 2008-11-21 11:41-0800 pw wrote:

> Hello,
>
> I keep encountering people who upgrade and upgrade
> and then have problems...etc etc.
>
> This 'syndrome' is not isolated to casual users
> either. There are loads of techs that seem to feel the
> need to continuously spend their lives in *upgrade hell*.
>
> I can see following proper security procedures and
> putting security updates on system but then...
>
> Does anyone have any stats that compare the productivity
> of:
>
> "If it ain't broke don't fix it"
>
> vs.
>
> "OMG! The 2nd decimal place of the version has changed!"
>
> ??
>
> When is a reasonable time to upgrade a system?
> Should we use a version until it's D.E.A.D. ?

Thanks, Peter, for your most interesting question.

The answer is (of course) "it depends".

On the issue of upgrade costs, if you are careful with your complete Linux
distribution upgrades (e.g., backup your personal data and take good notes
about exactly what you do so you can easily back out of it or start over and
refer to old notes for a new upgrade), there isn't any such thing as
"upgrade hell". I am one of those careful types so I rarely have trouble
with the rare complete system upgrades that I do. However, there is a subset
of experienced Linux users who are constitutionally incapable of being
careful so they constantly get into trouble not only with complete system
upgrades but also simple upgrades of individual packages.  So the cost of
upgrades depends strongly on what type of Linux user you are.

On the issue of upgrade benefits, that was an easy issue 10 years ago
because virtually all Linux software packages were getting rapidly better,
and you couldn't afford not to upgrade.  Now, that the Linux kernel and
desktop has mostly matured, it very much depends on what component you are
talking about.

* From the security perspective I think you almost always want a recent
kernel in the 2.6.x series.  There is an industry-wide kernel team lead by
Greg KH that does security updates to the stable 2.6.x series.  However,
from http://lwn.net/Articles/306596/, this last update is all that will be
done for 2.6.25 and 2.6.26 by this team so 2.6.27 will soon be the minimum
supported version of the kernel by this team.  Of course, various Linux distro
vendors try to security support older kernels for a while, but I am not sure
they are on top of problems for old kernels nearly as well as the team lead
by Greg KH for modern kernels.  I run Debian Lenny (testing) in part because I am
concerned about the security for older Kernels in Debian stable.

* X is a special case right now.  It is going through a tremendous period of
change with many aspects being completely refactored/rewritten such as mode
setting going into the kernel, GEM, DRI2, etc.  So struggle with a nearly
cutting-edge version of X (which is what I do by using the Debian unstable
version motivated by trying to help out by bug reporting problems for that
version) or deliberately stick with an old version until the newly factored
version settles down.

* KDE is another special case.  3.5 was great to start, but I am noticing
bugs are beginning to creep in (perhaps because of the above X changes) that
are not being addressed by developers because they are working so hard on
4.1.  I plan to move to KDE 4.x just as soon as it becomes available in
Debian testing.

Computers are not toasters.  Businessmen who use computers often don't
understand this basic fact so they get sold on some "enterprise" version of
Linux which promises to remain essentially unchanged.  All that means is
they are paying a lot of money to get access to old, buggy software which
has some but not all bugs being addressed by programmers who would probably
prefer to work on something newer and more relevant.  It also means the
upgrade cycle will have huge impact your business when it does come. My own
feeling is it will be more cost effective for businesses and individuals to
(carefully!) upgrade their distro each time a new version comes out (every 6
months if they are following Ubuntu).  The amount of change they have to
deal with then becomes managable compared to (say) upgrading every 3 years
or so.

My $0.02.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the libLASi project (unifont.org/lasi); the Loads of
Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________