http://disqus.com/by/quelgeek/enThu, 15 Oct 2009 04:16:55 -0000http://blog.tonybain.com/tony_bain/2009/09/the-rdbms-is-doomed-yada-yada-yada.html#comment-20107756<p>Your response shows we actually largely agree: the physical implementation, be it records or columns or even instantiations of objects, is irrelevant. Where we seem to disagree is that you either think the physical implementation needs to be made visible for some reason, or that there is a simpler, more flexible, and powerful way to think about the data than as sets of propositions.</p><p>Logic has served us well for a couple of thousand years. Set theory has worked well for a few hundred. You'd have to be very bold or very rash to think there's anything better on the horizon. Now, whether there is money to be made while people figure out there isn't, well, that's another matter.</p>quelgeekThu, 15 Oct 2009 04:16:55 -0000http://blog.tonybain.com/tony_bain/2009/09/the-rdbms-is-doomed-yada-yada-yada.html#comment-19486950<p>You confuse SQL DBMSs with relational DBMSs. You also seem not to understand the fundamentals of either, and more remarkably it seems Michael Stonebraker doesn't either--unless you also failed to understand him and put nonsense in his mouth.</p><p>There just flat-out is no viable alternative to the relational model. Until someone comes up with a better idea than set theory and first order predicate logic, any alternative is going to be worse. After 30-odd years there is ample evidence to support this claim. SQL has endured because it is pretty close to this approach and nothing radically different has been remotely successful for anything like as long. I could certainly wish that SQL were more faithful to the relational model but that's not important to enough people.</p><p>The only problem that really needs to be solved is how to get SQL DBMS developers to get over the limitations of 1970s era hardware. Back then it made sense to store rows as records; it doesn't now and never will again. There is nothing in relational theory that prescribes rows-as-records, and if it makes more sense to arrange the data in other ways under the covers there is no theoretical objection to doing so. They also need to get over the idea that "scalar values" or "atomic values" include only the kinds of data types supported in hardware by 1960s era mainframes. The relational model has no problem at all with arbitrarily complex types; they can be any digital data at all. Let me repeat that: ANY DIGITAL DATA AT ALL.</p><p>All that matters is that we be allowed to think in terms of tables so we can write queries in terms of tables, because, at the risk of repetition, no one has come up with a simpler, more concise, or comprehensible approach. When they do THEN we can throw out the relational model. Anyone who thinks different is just wrong and time will prove it.</p>quelgeekThu, 08 Oct 2009 05:45:20 -0000