Did IBM Hit A Nerve?

As IBM moves into the next phase of its DB2 Viper launch, Oracle seems to be getting touchy about IBM's agenda-setting around "native XML storage." From Barbara Darrow of CRN's unblog:

"How we store XML on the database is, excuse me, none of your business. The point is you can write an app using XML standards," said Mark Drake, manager of product management for XML technology for the Redwood Shores, Calif. vendor.

"Whether we shred it, parse it, it doesn't matter. There is no such thing as a native XML storage model, there is no W3c standard or 11th stone tablet, telling us how," he noted.

I actually concur with Oracle that "native XML storage" is a pure marketing-ism. As mentioned in this post, the relational model specifies nothing about storage. Tables are not stored tabularly on disk, and even if they were, I'd agree with Mr. Drake -- as a user, it's none of your business.

The social contract with a relational database is you give it (1) a schema of tables and columns, (2) data to load in that schema, and (3) then you get to run arbitrary SQL statements against that schema with defined, guaranteed semantics. How the purveyors of the DBMS actually do that is, indeed, entirely up to them.

I agree with Stephen Buxton, formerly of Oracle, now of Mark Logic, and co-author of Querying XML, when he says: "there is no storage, only representation."

How you choose to represent XML in an relational database, if indeed you decide to put it there at all, does however have ramifications on both functionality and performance.

As a simple example, if you choose to shred XML to a relational schema, then you can only work with XML content that is 100% regular and 100% conformant to a single schema. While that may work just fine for 10,000 XML-wrapped purchase orders (which probably originated from an relational database in the first place and got published as XML for messaging purposes), it does not work in the subtle and ambiguous world of document content, where things are not so regular and not so normal.

If you choose to put XML into a column of type BLOB, there are other ramifications. As there are with putting it into a column of type XML, as DB2 Viper provides.

But without drilling into the pros and cons of these alternatives (or those of using a native XML content server like MarkLogic), let me observe the softer side of this issue. It appears that I'm not the only me who's noticed that IBM's XML marketing is pretty loud and pretty effective.

For another view, see Curt Monash's blog on this topic, entitled "Oracle Getting Touchy About XML."