Oracle Broadens Support for Open-source R Analytics (PCWorld)

Posted Aug 26, 2012 16:00 UTC (Sun) by njs (subscriber, #40338)
In reply to: Oracle Broadens Support for Open-source R Analytics (PCWorld) by mjg59
Parent article: Oracle Broadens Support for Open-source R Analytics (PCWorld)

Right -- that's why I didn't mention Sunperf, because I don't know enough about its licensing and how it fits into the Solaris ecosystem.

IIUC the situation is legally identical to the one around binary NVidia drivers on Linux: these math libraries implement standard interfaces (math.h, BLAS, LAPACK) so everyone knows that R itself is not a derived work of MkL/ACML or vice-versa. If you want to combine them in the privacy of your own home, no problem. But distributing a combined package of R-built-against-MKL/ACML seems very dicey.

Oracle Broadens Support for Open-source R Analytics (PCWorld)

Posted Aug 27, 2012 14:48 UTC (Mon) by pboddie (guest, #50784) [Link]

Indeed. As always, with GPL-licensed works, it helps to think of the running system and then the act of the user exercising the right to see and modify (and distribute) the code for that complete system (down to a certain level as noted in the licence).

If there's a portion of the code that is proprietary and this was bolted on (using dynamic linking, perhaps) under the user's own initiative, then they have chosen to be able to exercise their rights for only the part that has the copyleft licensing: no-one has made any promise to them about the proprietary part of the resulting system.

On the other hand, if the proprietary code was provided and is loaded into the system under someone else's initiative, that party has committed to letting the user exercise their rights for the whole system. If that other party withholds those rights for the proprietary code then they have violated the licence of the GPL-licensed work by distributing it in such a configuration.

Oracle Broadens Support for Open-source R Analytics (PCWorld)

Posted Aug 27, 2012 18:35 UTC (Mon) by dashesy (guest, #74652) [Link]

OpenCV uses a similar idiom, if IPP is installed it can be used for better performance, otherwise it resorts to the open source version. The additional speed of OpenCV with IPP, or numpy with MKL is very dramatic for certain applications.

I personally use numpy built with MKL because that is the only way to beat MATLAB performance with matrices. I also prefer numpy to R (which I have little experience with), but if Oracle is going to distribute the MKL version of R binaries, it is just something to thank them for. Not everybody needs to build every single package, and once the MKL is installed one can recompile binaries, and benefit from MKL if a similar approach to OpenCV is used.