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

[Posted August 24, 2012 by n8willis]

At PCWorld, Chris Kanaracus is reporting that Oracle has announced several new initiatives supporting the open source statistical analysis tool R. Ports to Solaris and AIX are among the new offerings, as is compatibility with some additional tools. "Oracle will deliver an open-source distribution of R that can work with the Sunperf, MKL and ACML math libraries, which are provided by Oracle, Intel and AMD, respectively. This will allow R 'to run extremely fast on all compatible hardware,' Oracle said in a statement".

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

Posted Aug 25, 2012 20:36 UTC (Sat) by reddit (guest, #86331) [Link] (1 responses)

Oracle doing something good for others?

Where's the catch?

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

Posted Aug 27, 2012 15:20 UTC (Mon) by drag (guest, #31333) [Link]

R language is used commonly by scientists and it runs on clusters.

Apparently Oracle is interested in selling hardware, software, and services to that group of people.

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

Posted Aug 26, 2012 0:20 UTC (Sun) by njs (subscriber, #40338) [Link] (11 responses)

As far as I can tell, distributing a copy of R compiled against a closed source math library like MKL or ACML is a violation of R's license (GPL). This hasn't stopped Revolution Analytics, who've been doing exactly that for a while, so maybe I've missed something, but it sure seems sketchy to me. I'd like to see more details on what exactly Oracle's lawyers think they can do here.

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

Posted Aug 26, 2012 2:16 UTC (Sun) by Ed_L. (guest, #24287) [Link] (6 responses)

Not sure what is your issue. Back in the day, GNU tools ran on all sorts of proprietary systems. Still do. AIX, HPUX, Solaris....

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

Posted Aug 26, 2012 4:59 UTC (Sun) by spaetz (guest, #32870) [Link] (5 responses)

running a program on a proprieary patfrm is diferent from linking your gpl app to a propretary library to be used in the program. Which is what hey seem to be doing.

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

Posted Aug 26, 2012 8:51 UTC (Sun) by Ed_L. (guest, #24287) [Link] (4 responses)

How is linking your GPL app to a set of proprietary OS system libraries different from linking your GPL app to proprietary replacements/enhancements to those system libraries? Isn't a proprietary math library pretty much a proprietary math library, and the relevant license the proprietary license under which the proprietary math library is distributed?

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

Posted Aug 26, 2012 9:15 UTC (Sun) by mjg59 (subscriber, #23239) [Link] (3 responses)

Because the GPL explicitly provides an exception for proprietary system libraries.

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

Posted Aug 26, 2012 16:00 UTC (Sun) by njs (subscriber, #40338) [Link] (2 responses)

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.

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

Posted Aug 26, 2012 16:27 UTC (Sun) by butlerm (subscriber, #13312) [Link] (3 responses)

>As far as I can tell, distributing a copy of R compiled against a closed source math library like MKL or ACML is a violation of R's license (GPL).

The normal way around that would be to create an open source library that implements the same ABI as the proprietary one, compile against the open source header files, and then allow the end user to choose which implementation to load at runtime.

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

Posted Aug 27, 2012 17:02 UTC (Mon) by drag (guest, #31333) [Link] (2 responses)

Well header files should not be copyrightable so it doesn't matter which ones you compile the application against.

The GPL only covers derived works. Whether or not something is 'derived works' is not a question of the software, the license, or common sense. It's a question that only a Judge can really answer.

Regardless it's very unlikely that a GPL software will be considered a derived work of a closed library using a public API.. even if there wasn't a exception for it in the license. There are exceptions to this, but it's going to be a case-by-case basis and it's doubtful you could draw any generalized conclusions.

Not unless a Judge decides to make a sweeping change to how derived work is decided by changing the precedent.

This is USA stuff. Mileage will vary drastically based on country in question.

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

Posted Aug 27, 2012 22:34 UTC (Mon) by nix (subscriber, #2304) [Link] (1 responses)

Header files should not be copyrightable? Really? Function prototypes, perhaps, but you do realise that some headers can contain really quite a lot of actual code? (Look at the libstdc++ headers sometime, or the obstack headers, or, hell, bits/*inline*.h from your installed glibc. If *that* isn't copyrightable, *no* code is copyrightable.)

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

Posted Aug 27, 2012 22:40 UTC (Mon) by Cyberax (✭ supporter ✭, #52523) [Link]

Header files shouldn't be copyrightable when they contain no functionality or only trivial functionality.

Complex header files (a-la Boost header-only libraries), of course, are copyrightable.

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

Posted Aug 26, 2012 16:32 UTC (Sun) by jengelh (guest, #33263) [Link]

Thinking of it, if such math suites like R, Maxima, etc. could use a GPU (or even a set of GPUs), *that* would really help people get their work done fast. [Nobody likes writing the low-level code, e.g. with CUDA, oneself.]

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

Posted Aug 27, 2012 15:26 UTC (Mon) by itoen (guest, #86424) [Link]

It's debatable whether linking against MKL and ACML will make much of a difference for R programs that are not spending their majority of time in matrix multiplication or matrix decompositions (ie. non-trivial code that does a lot of data manipulation).

A big problem with R is that it's an interpreted language. There is a "compiler" for it, but it merely compiles the code into form that is a bit faster to interpret. The issue is so pronounced that elaborate bridges between R and C++ have started to become popular, eg. the Rcpp package, allowing bidirectional links with C++ matrix libraries such as Armadillo.

If Oracle is serious about speeding up R, they should really look into making a JIT for R, rather than making noise about trivialities such as swapping out standard LAPACK for MKL.