[linux-elitists] proprietary academicware, was [Beowulf] Please help to setup Beowulf

D. Joe Anderson deejoe@etrumeus.com
Fri Feb 27 05:56:23 PST 2009


On Tue, Feb 17, 2009 at 04:07:38PM +0100, Eugen Leitl wrote:
> ----- Forwarded message from "Robert G. Brown" <rgb@phy.duke.edu> -----
> 
> On Tue, 17 Feb 2009, Mike Davis wrote:
> 
> >Geoff Jacobs wrote:
> >>
> >>Why do you say this? Debian (5.0, what I checked just now) includes all
> >>the basic Beowulf elements and even tosses in GROMACS pre-compiled for
> >>OpenMPI. If it's what his people are used to there's no reason to switch.
> >>
> >I don't know about RGB, but I would argue that you let your applications 
> >lead

> >All of his work and the various programs that interested him required 
> >fortran which was still another update. After several evenings of searching 
> >and trying to setup proxy servers for updates to nodes. I switched the 
> >entire cluster (4 nodes) to CentOS and had his primary application (GAMESS) 
> >running in a few minutes compiled for both serial and ddi sockets.

ah, GAMESS.

There's a signature difference in this discussion between
GROMACS, which is free enough that distributions can include it,
and GAMESS, which is far from it:

    The license acknowledges the Gordon group's exclusive
    distribution rights to GAMESS, and also prohibits you from
    making copies of the GAMESS code for any purpose except use
    at your own institution. Please note the distinction between
    "a site license at no cost" and terms such as "public
    domain" or "freeware" or "open source".

  http://www.msg.ameslab.gov/GAMESS/download.html

Why is it that, if you actively seek out licenses which stay out
of your way, you are a license zealot, but if you stick with
free-as-in-pizza* licensing like so much of this kind of
"academicware" (instead of, at worst, using
GPL-plus-proprietary-derivatives dual licensing), you aren't?

Academicware is just free enough to undersell more commercial,
but still proprietary, competitors and to suppress most of the
"we must replace this because it's a PITA" itch.  It doesn't get
rid of the underlying cause of that sort of itch, and still
remains a form of rent-seeking.  But it pushes the itch far
enough away that people don't recognize the itch as coming from
the application license.

Instead, as is done here, the problem gets recast in terms of OS
and distribution choice, and what must be done to accomodate the
application's restrictive licensing.

In a way, it parallels the "drivers: in, or out, of the kernel?"
issue.  If your driver is in the kernel, everybody wins, you and
your users get help with your driver from the likes of Greg.  If
your application is in the distribution, everybody wins, you and
your users get help from the package maintainer.

If, for either drivers or applications, it isn't free software,
it doesn't matter that it is free-as-in-pizza, because usability
still suffers.

> >In short, the applications are the whole point of a working cluster. If an 
> >OS will not readily support the apps, you need a new OS.
> 
> Absolutely.  And as soon as we find out what he wants to do and what
> he's trying to do it on, we'll figure out if he needs a new OS.  But he
> probably CAN get by with Debian OR Centos OR Fedora OR Oscarocks OR SL
> OR SuSE -- with only mildly differential work.  So we need to eventually
> hear how skilled he is with any of the distros.  

* It occured to me, after seeing my hojillionth campus flier for
a student group interest meeting that advertized "free" pizza,
that at least some of the befuddlement the "free beer"
terminology embodies is generational.  "Free beer" might have
been something people reasonably spoke of in a less litigious
time, when the drinking age in the US included, rather than
excluded, a majority of the college-age population.

-- 
D. Joe



More information about the linux-elitists mailing list