Lahey Licensing of Fortran compiler for Linux - in detail ;-)

[Christoph Best]

Craig Tierney writes:
 > On Tue, Jan 21, 2003 at 02:57:06PM -0500, Mark Hahn wrote:
 > > it makes a kind of weird sense: if Lahey has to work harder to make their
 > > runtime scale to >64 CPUs, they want to be paid more.
 > 
 > You are correct IF their compiler is aware of the cluster runtime enviroment
 > then the price should reflect that.  Are they providing their own MPI or PVM?  
 > Are they providing a version of HPF?  No.  So compiling for a single CPU is
 > no different than for a 1024 cpu MPI process. Scaling is an issue of the developer
 > not the compiler engineer.
 > 
 > Portland Group doesn't put a restriction how you use their Fortran compiler.  They
 > do put a restriction on their cluster tools.  Those you have to license by CPU.
 > If you want to use their HPF then you need to purchase a license for the number of
 > CPUs you want to use.  That makes sense for the effort they put into their compiler.
 > Portland Group does not charge me if I want to link my source against a free version
 > of MPI and run on multiple cpus.
 > 
 > I am not against Lahey's licensing policy.  They can do whatever they want its their
 > product.  When I talked to one of their reps at SC2001 and told them that no one else licenses 
 > their compiler this way they didn't believe me.  

Actually, the thing that would be worrying me is that, as a compiler
vendor, Lahey doesn't seem to completely understand what a "cluster"
is. Even 10 years ago, our University's compute service was a
"workstation cluster" with about a hundred CPUs - well, 100
workstations in public access rooms - connected by Ethernet, and you
theoretically could even have PVM running. Lahey doesn't say in their
license that the programs are required to communicate, or even to run
in parallel, they just say that you cannot run on a cluster with more
than 4 CPUs.

OK, they probably do not mean that, but lawyers (and courts) do not
like to think about who means what.  This stuff sounds like Marketing
has some confusion about the difference between a SMP and a Beowulf
cluster.

Talking about legalities, another worry in the wording of the license
is that - at least from the fragment Andrew Leahy posted - you cannot
sell or give away the compiled code. They grant "you", the buyer,
explicitly the right to run the executable, and that might indicate
that anyone else who wants to run your code would have to get a
license from Lahey, too. Again, they might not mean that, but the
compiler license should not restrict how you can use the compiled
code.

It sounds like they just added these two sentences about clusters to
the license to milk Linux clusters without thinking about the legal
implications.

-Christoph

-- 
Christoph Best                                  c.best at computationalgroup.com
COMPUTATIONAL science&consulting            http://www.computationalgroup.com
Frankfurt am Main, Germany