Thursday, April 15, 2010

A few notes from the History of Database Systems BoF

I will not write much on what went on at this BoF, but a few words are appropriate I guess.
  • We were some 10 to 15 people in the room at the end.
  • I started the thing by talking about the ancient history of databases, and went on to talk a bit on the reasons of the Relational databases appeared on the scene.
  • We talked more on these reasons. The old Network and Hierarchical were largely tape oriented, even when data was on disk.
  • My theory on search being a very major factor for adopting the relational database technology in the 1970s and on seemed to be accepted.
  • As I went on to discuss search in an RDBMS being contextual, and the need for non-contextual search caused quite a few debates.
  • That non-contextual search will be a factor in moving to NoSQL, as is a theory of mine, was not accepted by anyone else but myself :-)
  • Contextual and non-contextual serach means, and if this even is search and what these terms mean was discussed in gruesome detail.
  • This brought on a NoSQL debate that lasted till the end of the BoF.
  • That NoSQL is about performance was largely accepted.
  • That key-value storage is a key behind performance was not (and I sure don't see it that way).
  • The value of a Key-Value store in itself was discussed in detail. Are we just storing any kind unstructured data as a Value, or is it XML (which is most the case currently) or an Instance of an object.
  • We alsodebated the storing on a set, as in an RDBMS, vs. an Instance, as a key-value store may be seen, was also debated. And if an RDBMS really is set-oriented and if a K-V store stores an instance was a hot topic.
  • I think we eneded up with a notion that we will probably see a mix of RDBMS and K-V stores in the future, that they are complementary in the short to mid-term.
  • In the long terms, I claimed that a K-V store will not persist as a generic solution, as it actually has less functionality than an RDBMS, whereas other claimed that a K-V store applied properly with instances and instance pointers ´within values is the way to go.
  • Whatever happens, it will be interesting.
Thank you everyone for participating in the debate, it was enlightening to me, and I hope you at least had some fun and learnt something also. And I hope I didn't take up too much space myself in this debate.

/Karlsson

1 comment: