Hi

Have you looked at www.Matisse.com

Simon

>I was reading some Microsoft .NET propaganda and thinking of the implications of distributed OOP on the classical model of a database. Databases, or more specifically tables, have been around for a long time and have been a staple of Information Technology departments everywhere. But are we witnessing the end of describing data by rows and columns?
>
>Much of the effort in database design is spent modeling real-world objects as rows in tables. When the real-world objects are sufficiently complex, we use more than one table to describe it. We apply the rules of database normalization to ensure that the model is effiecient and extensible. But is this all really necessary?
>
>Imagine for a moment a world without rows and tables. All we have are objects with their properties, events and methods. Could we build a system that stores all of the information that is currently contained in rows and tables as objects? Instead of breaking up tables to build the proper relations and atomicity, could we not create a model that directly mimics the objects instead?
>
>It now seems that all of the requisite pieces are in place to break away from the database as we know it and move towards a completely object-oriented way of storing and retrieving information. The only question is whether this would be better than the way it is now.
>
>At an abstract level, a table is identical to a collection of objects. We index tables for quick retrieval of data. Could we index collections of objects in a similar way? In that same light, could we efficiently retrieve large quantities of objects as quickly as we can retrieve large quantities of rows?
>
>Add to that the natural extension of collections of objects to distributed storage. I see this with .NET. If .NET takes off, we will be able to design distributed collections of objects that store information and that already have all the code needed to process the information. Will this eventually fufill it's potential and replace our "classical" databases once and for all?
>
>I can picture a request for information originating from a workstation and being sent out across the network. Similar to an IP packet, this request would be wrapped with information that would allow routers to forward the request towards its intended destination. Once the request has found the collection that contains the information it is looking for, it would attach the object containing the information to itself and begin the second leg of the trip back to the workstation.
>
>Of course, these ideas are not new to Object Oriented Programming. But we create programs where everthing is objected oriented except the data. Then we create an additional layer of code that transforms the non-object oriented data into an object oriented form for use in our code. Will we be able to remove that layer of complexity from our system designs?
>
>Finally, there are trade-offs to be considered. Normalized databases are extremely efficient in terms of the storage space required to hold the information. It is likely that a pure object oriented database will not be as efficient as a normalized database. However, we have already made those trade-offs in our OO code today. No one can argue that the change from structured programming to OOP brought with it less efficient code in terms of instructions and storage space. Those trade-offs are made because OOP is enormously more flexible than structured programming. Are we willing to make similar trade-offs in our databases?
>
>- Keith Payne