Thursday, October 27, 2005

RDBMS RIP?

At face value this article would appear to be, at best, sensationalist. But if you delve below the hyperbole, the essence of Gartner’s view is that data will become further decentralised and distributed. Unfortunately, the majority of the article weakens this argument by focusing on some simplistic RFID scenarios that do not withstand further investigation.

Consider the can of soup example with the inventory system that has no historic reference data. Say the inventory system determines that there are 100 cans of soup at one point in time, but later on discovers that there only 95 cans. Does it even know that 5 cans have gone since the last check? Have 5 cans been sold? What if 2 were stolen? What if a customer purchased a can, left the shop briefly, and later called back still carrying the same can of soup? How would the inventory system recognise that possibility if it accidentally discovered the can? Will the customer get charged again when he leaves the shop? Forget the soup, what we have here is a can of worms.

All of the above problems ostensibly come from the lack of any historic reference and yet the article only very briefly mentions that this still remains an essential aspect. This brief mention appears in the final paragraph, after the previous claims have left you in a state of outrage.

The fact is, RFID tag scanners and processes will produce event data essential to the understanding of the distribution system and its efficiency. Moreover, with tagging on individual items, there is likely to be terabytes of the stuff flowing throughout any reasonably large enterprise. You will not want to, or need to, pour all of this event data into a central database. I think Gartner is promoting the idea that you keep the data close to its origination and distribute it. This is more an argument for EII and SOA than an argument for the demise of databases per se. If nothing else, you need to retain historical data for manageability, security and accountability. But it doesn’t need to be centralised and detailed immutable historic event data doesn’t need to be shoehorned into a RDBMS as you only need to access it for search and aggregate it for analytics.

However, I do dispute the claim that XML is unstructured data... more comments about that later...

0 Comments:

Post a Comment

<< Home