Back to build, not buy

Packaged analytics 'do not always scale'

Analytics is no longer an obscure "black art" perpetrated by pointy-headed rocket scientists for whom the term "applications developer" is totally inadequate. It is now a big part of everyday production environments and is getting bigger and more important.

The number of factors that need careful analysis when running a major business venture – particularly in areas of fast-moving financial management, risk analysis, and business intelligence – are growing every day and moving increasingly towards the centre stage of business management.

More and more applications developers therefore need to be able to work in this field. That, however, is somewhat easier to say than to achieve.

According to Visual Numerics CEO Phil Fraher there are a number of factors that now impede the rapid development of high-performance analytical applications and services, and which in turn are bringing about a subtle but significant change in the way that such applications will be developed.

Visual Numerics has operated in the analytics field for some 30 years, and has built up a large selection of statistical algorithms encapsulated into downloadable modules. These make up the company's IMSL libraries and are already widely used by specialist developers working in the analytical field. Historically, the libraries have been available in Fortran and C.

Recently, however, the company has ported the libraries to two far more widely used environments, Java and .NET and added functionality in the form of a Dense Linear Programming Optimiser aimed at analytics in the BI and data mining sectors.

The ports to Java and .NET have had no impact on the performance of the libraries, according to Fraher, who claims the new versions can run LSSOL - which is designed to solve a class of linear and quadratic programming problems – as fast as the existing Fortran-based versions.

With a large number of developers now working with Java and .NET, this is opening up new opportunities not only for the company, but also for those developers seeking ways of addressing complex analytical problems.

"We are finding most developers are migrating to Java, and with more universities teaching Java there are a now more developers appearing with those skills," he said, adding that Java also had the advantage of being platform independent.

This opens up the potential for developers to use the libraries at a time when both the demand for complex analytics is growing and, in Fraher's view, the limitations of packaged analytical solutions such as the SAS System are beginning to show up.

"These packages do not always scale in the way needed by critical applications, so we are seeing a move towards developers building them in Java, using the IMSL libraries, rather than buying a packaged solution."

The traditional problem with building applications is that it takes far longer to achieve a result than buying and adapting a packaged solution. Fraher, however, suggests that using the library modules means that complex analytical applications can be built in a reasonable timescale. "And the important competitive analysis applications are not general purpose in nature, so they really have to be built."

The company is now looking to extend the scope of the libraries, with data mining and forecasting markets being particularly targeted, in addition to the provision of more finance algorithms.

Fraher wants the company to provide a comprehensive set of building blocks so that modern developers can still build applications using complex mathematics. "In the old days the PhDs could do the business and the programming," he said. "But now they can't. So using the libraries as building blocks is now an important tool." ®