Relational database pioneer weighs in on NoSQL

That comment was about the NoSQL community's hostile reaction to his criticisms of MapReduce, and it does indeed show that they're not interested in knowing about databases, because they read his criticism in the context of an imaginary (or only-real-because-people-think-it's-real) SQL vs. NoSQL struggle. Instead of seeing it as an analysis of the shortcomings of MapReduce (and being curious about the perspective or needs that might make those shortcomings important,) their simplistic reaction was that he was fighting for the wrong side and needed to be discredited.

One of the comments on the blog exemplifies the reaction perfectly:

Technically, I agree with much of this article ... <snip> ... But none of that matters. It’s all about hearts and minds, and if the DB industry adopts the attitude in this article, it’s back to what I said at HPTS back in 2001: We Lose.

Screw content, screw technical judgment, it's about attitude and "winning" the struggle "against" existing database technologies.

He knows how to get press, and playing up that confrontation yields links and references. Just as many NoSQL pieces start off with some often ridiculous attacks on YeSQL.

I'm still trying to understand how SciDB differs from VoltDB. We're currently evaluating VoltDB for some analysis processes.

I would agree that MapReduce and NoSQL are different things, but so are NoSQL and AJAX.

I've always thought of NoSQL as a class of durable, distributed lookup tables, and the style of programming which uses such tools as a primary data store. It's basically bdb scaled up with nice libraries built on top. Document databases seem to me to be api sugar.

Indeed you can - I've just written a very simple MapReduce framework that sites on top of the SQL database inside Safari on the iPhone. At the moment the JSON documents it uses are stored in a SQL table, but you could convert a more "traditional" table structure into JSON dynamically.

Agreed; in particular, the quotes read like poorly-transcribed scribbles that don't even form sentences.

They aren't diametrically opposed. They are different aspects of the same thing.

At the end of the day, it's all just tables. NoSQL gives you one table and all sorts of scalability and robustness and tools for working with that table. RDBMSs give you many tables, and all sorts of robustness and scalability and tools for working with those many tables, and for putting them together in interesting ways to get answers to your questions.

NoSQL allows for much greater scalability for your one table because it doesn't have to assume you are doing anything too complicated with it. Usually, you just want a few records at a time. That can be made fast, even if there are billions or trillions of other records, and they do make it fast.

RDBMSs trade that sort of scalability for the ability to easily ask very very complicated questions and get answers reasonably quickly.

And here's this guy in the article talking about RDBMSs with column oriented tables instead of row oriented tables. And then he bashed MapReduce, which is about transforming data, and has nothing to do with tables or storing and retrieving data. So NoSQL and RDBMSs are not diametrically opposed. Talking about them is like comparing apples and oranges, which are both fruit, but not equivalent. Throwing MapReduce in there is like comparing apples and oranges and blenders.

He's qualified because he has deep domain expertise in data storage. It's interesting and valuable to hear opinionated points of view from people who have clearly demonstrated skill and ability in a field, even (especially) if they disagree with the consensus view.

Stonebreaker isn't a relational zealot. He's been inventing new stuff outside the original relational model for ages. Remember object relational databases? The relational purists hate this guy.

Seriously! I could adjust for some spelling errors, but sometimes the entire sentence lost it's meaning.