TeraSolve; IBM and BI – I take it all back

Written By:
Published:
Content Copyright © 2006 Bloor. All Rights Reserved.

Okay, I confess. I have been beating IBM up about how it needs to buy a BI player in order to be a credible player in the space and it turns out that I got it all wrong, because IBM already owns a BI player. Well, actually, a bit of a BI vendor. A very small bit. And I’m not actually sure that the Information Management people actually know about it but, nevertheless, IBM owns a very small (less than 5%) portion of a New Zealand based company called Descisys.

Now, despite the fact that the company has been around for years you probably haven’t heard of Descisys. However, last year the company raised some VC funding (having been privately funded previously) and re-branded itself as Information Edge, at the same time opening offices in the UK (where HM Treasury and Abbey National are customers) and the US.

So, what does Information Edge do that you might be interested in? Well, the company has some special-purpose analytic applications in the government and telco spaces but the most interesting thing is its TeraSolve product.

To put it simply, TeraSolve is ROLAP with dynamic (rather than static) cubes, which not only does more than ROLAP can do but does it, according to the company, ten times faster. However, where it is similar to ROLAP is that it still relies on the relational database for storage; it is just that it stores the data differently and more efficiently.

TeraSolve was originally designed to run on DB2 but has since been ported to both Oracle and SQL Server. It provides multi-dimensional functionality but without actually ever instantiating this. There are two parts to this: first, it separates the data from the metadata so that all the relationship information about dimensions and hierarchies is stored separately from the data. And secondly, the aggregated data that you want to retrieve is stored as BLOBS (binary large objects or, probably more accurately in this case, binary small objects). On the surface of it, this appears weird but TeraSolve includes its own indexing techniques for this data and optimisation to ensure maximum storage and retrieval capabilities.

In addition to performance, there are a lot of advantages to this approach. First, there is no sparsity issue: if there is no data in a cell it is not stored. Secondly, the back-up, failover and security facilities that normally apply to relational databases, and which are typically superior to OLAP, all directly apply to TeraSolve data. Thirdly, you can support any size of cube with any number of dimensions: having to support multiple cubes can be a real pain that limits scalability. Fourth, it is very easy and quick to add a new dimension, say. In contrast, it might take months to make such a change in a conventional environment. And, finally, because the metadata is separated from the data you can change either without affecting the other. For those of you familiar with Kalido, this is very similar: it means that the data is isolated from changes to corporate structures and, indeed, could be used to help test potential structures as well as maintaining a sort of corporate history.

You might think that was it: that’s a pretty good roll call of advantages. However, there is one more unique feature of TeraSolve, which is that it uses declarative processes for calculations. Other products use a procedural approach. Now, the problem with procedural methods is that the order in which you specify operations can make a difference to the result. Most people aren’t aware of this. It doesn’t mean that there are errors in the calculations your system makes but there may be. When, in proofs of concept, Information Edge has come up with different answers from the incumbent supplier it has to demonstrate that it is right and the previous methods were wrong—fortunately, it is able to do that.

By all these counts TeraSolve is an impressive product and it could transform the processing of aggregated queries. Indeed, it could be as disruptive a force for traditional ROLAP as appliances are in the wider data warehousing market. Given that IBM owns a (small) chunk of Information Edge it should take a closer interest.