I was planning to be done with all the posts about ICDE 2015 last weeks, but then life came my way, and I didn’t write any new posts for quite a while. Moving on…
One on the important events on of the conference was a ceremony of “10 Years most influential paper” award. This year it was given to Michael Stonebreaker and Ugur Cetintemel for the “One Size Fits All”: An Idea Whose Time Has Come and Gone – the paper presented at ICDE 2005
The paper can be found here.
Stonebreaker himself didn’t come to Seoul, but he recorded a video presentation, which can be viewed here (just FYI – very slow!), and after this presentation was played, he answered several questions from the audience via Skype.
To be honest, I have very mixed feelings about this presentation (not like Stonebreaker would care :)). I understand, that one might need to be controversial to attract the public attention to some issues, but I feel like he uses some terms incorrectly (or, let’s say, in non-conventional way). During his presentation he was saying that if 10 years ago it was “not one size fits all”, now “one size fits nothing”, and then he says, that what he means – that “traditional relational databases” are not good for anything in the modern world.
The question I have – what exactly is meant by “traditional”? I think, the database is either relational, or not relational :). If you listen closely to what he means, you will figure out, that by “traditional” databases he means big companies, whom he call “elephants”, i.e. Oracle and MS SQL Server, and DB2.
I disagree, when he says that the “column storage dbs” are something different, or that in-memory dbs are something different. In both of those (and many other cases) the theoretical foundation of the databases in question are still the same – they are relational in their nature. I think, that this fact actually proves, how versatile the RDBMS’s can be, and how may different “families” of them can exist.
Also, when he says, for example, the queue-based systems are so much faster than the “traditional” databases, I also feel that I’ve been cheated on – he does not tell, how the data is persisted, he does not elaborate, how the systems deal with complexity, and most importantly – even if writing is indeed faster – how the search is? Is it faster, or the same, or “nobody bothered to measure”? (I sincerely hope it’s not the latter case, though!). But in any case, I would like to hear about this kind of comparison.
For those of you who never came across this paper I highly recommend to read it. As my husband pointed it out, “the most influential” does not mean it is right, it means, that people talked a lot about it – and we should learn from it, whether we agree or disagree.