Nice to see a story about bloody software patents which aren't about someone being a massive shithead. Unusual to hear this, but well done, Google!
Three months after securing a patent for MapReduce - the distributed number-crunching platform that underpins its world-spanning infrastructure - Google has granted a license to Apache Hadoop, easing infringement concerns hovering over the MapReduce-mimicking open source project. Apache legal counsel Lawrence Rosen announced …
"The map and reduce functions described by Google have been a part of parallel programming for decades."
So that sort of rules out novelty and implies existing prior art.
Patenting something that others have been using for "decades"? Sounds pretty, er, shitheaded to me, regardless of what happens to the patent subsequently. "We own your sorry arse, but we'll let you use it for free" still means that your arse belongs to someone else.
And statements similar to "google have always played nice with their patents so that's all right" remind me of the standard boilerplate disclaimer on financial products: "Past performance is no guarantee of future returns". Google is a big company and therefore subject to phrases such as: "boardroom coup", "strategic reevalution", "monetise our existing asset base" and other such nasties, just as much as all the other elephants jockeying for position in the room.
On the one hand, I can't say this is exactly good news for Open Source (OS) projects since all it really means for Google to have given a "free pass" to one high-profile OS project is that they've evaluated the pros and cons of either suing or automatically granting a license and decided that granting a license garners better monetary returns via PR than suing does. It does nothing to help any other OS project with a lesser profile. Also, the Apache license contains no "patent non-compete" clauses, so this announcement is of questionable value to downstream projects.
Which brings me to the second point... basically the ridiculousness of software patents in general. It's quite clear that the "map-reduce" principle has been well known in computer science since, well, pretty much forever. I enter in as evidence here the well-known "mergesort" algorithm. Map-reduce is nothing more than a generalisation of the basic idea of mergesort: ie, partition the problem space, have each partition operated on individually, and have a (usually O(n)) "reduction" step when all partitions have been processed.
It's a sad day when a generalisation of a well-known technique can get patent protection. This both from the point of view of mathematical abstractions (which shouldn't be patentable in general, and generalisations doubly so) and basic computer science (where algorithms in general shouldn't be patentable, and even if a case could be made for them, they'd have to be *more* specific, not less). Both are obviously flip sides of the same coin.
And one more point. All the previous commenters are idiots. Or possibly shills, but I'd prefer to apply Hanlon's Razor, obviously.
I wouldn't assume Microsoft Bing is using Google technology for Bing. A quick review of the Microsoft Research web site will find a compelling technology and programming environment known as Dryad and Dryad LINQ ( language inline native query ) that offers another way to get the functionality many people find useful in Hadoop.
I think the comment in the article relative to the use of Hadoop in Bing is a bit of flame bait and opinion, not a fact.
Biting the hand that feeds IT © 1998–2020