A lot of people are unaware that Ingres in the second largest open-source vendor (after Red Hat). Its core offering is the well-proven Ingres relational database, but today Ingres announced an interesting new direction. Called VectorWise, the new technology is a columnar database kernel aimed at rapid performance for query processing (it still uses the Ingres database optimiser). Ingres have been working with Intel and the CWI of Amsterdam on the new project, which is designed to maximise the performance of query processing on modern hardware chips, such as the new Intel Xeon 5500. Existing Ingres customers will be able to take advantage of this new facility for analytic processing.
The company released some nice-looking statistics regarding performance of queries compared to conventional relational storage, though in some sense this is not surprising since columnar approaches are inherently more efficient than row-oriented databases at queries (conversely, row-oriented databases win at transaction processing). What is more interesting is the collaboration with Intel, and the fact that the database kernel is open source, with the pricing advantages that this brings.
The new product seems to squarely play in the data warehouse appliance space, along with other appliance vendors such as Netezza and Greenplum (the latter is based on the open source Postgres database). Many of these appliance vendors also use a columnar database e.g. Sybase IQ and Vertica. For this reason, it would be interesting to see performance benchmarks both of the "official" TPC-H style (which are independently verified) to compare VectorWise performance not only against conventional relational, but also against other appliances, which is realistically where it is going to play in the market.
Like any appliance, the key to wide adoption is not just blazing speed on an abstract benchmark, though. What customers will need to see is the business case for replacing existing data warehouses using more traditional approaches. For this it is not enough for the vendor to demonstrate hardware savings. This has to be offset against the migration effort involved (relational databases may appear to be based on a SQL standard, but in reality a lot of code is wrapped up in proprietary code, such as stored procedures and triggers). Ingres actually has a migration centre to help customers with such migrations, so it will be interesting to see some real case studies in due course in order to understand how the savings involved compare to the migration effort in reality.
Of course there are situations where issues other than cost reduction come into play. In some cases speed is the key factor, for certain business-critical analysis. There will also be circumstances where new applications will be possible to implement that were previously either impractical or too costly. A measure of the success of VectorWise will be in how widely it is taken up by the business application community. Certainly its open source nature will have appeal here, as vendors can avoid costly database licensing costs. At present this is a project, not a product, and it is unclear when the fruits of this project will make it into general release. This interesting new technology will continue the market momentum that specialist data warehouse appliances are making to cater for demanding analytic processing needs.