What if just about everything we assumed about building back-office applications in the enterprise turns out to be wrong?

That's a premise that a small company called N_gine, based in Highlands Ranch Col., is betting on as we start to enter a new age of parallel computing based on multi-core processors.

Led by William Hinkle, N_gine has a tool and set of processes for breaking down financial transactions into a set of more atomic-level components that can be truly processed in parallel. Hinkle argues that virtually all back-office applications today are bloated because they were designed to run in a serial fashion. That serial approach also puts a great deal of emphasis on storage systems as the engine for managing updates to these applications. In contrast, he notes that scientific applications have been running in parallel for years because they tend to be quantitative in nature. That means it's easier to break them down into a series of modular components that could be run across multiple compute engines in parallel.

Hinkle, who started his career at Citibank and has previously led two small companies--Dynasoft and Keyvest--in the transaction processing space, is now applying those same basic concepts to back-office enterprise applications. Instead of processing a series of what-if statements in a serial fashion in order to figure out what action to perform next, N_gine allows a developer to break the application down into a series of modular functions that can be performed simultaneously across multiple processors.

A key element of this solution is a process known as parameterization, which basically breaks down the code associated with financial transaction processes into more atomic sets of data. The person running the application then updates the system by adding or subtracting an input value to a particular disk location.

Hinkle says this approach will allow users to decompose a financial transaction to run in parallel with, for example, a performance of about 280,000 transactions in under 50 minutes on a machine that would cost probably less than $10,000. More surprisingly, the N_gine server only needs 2GB of memory to run.

The implications this could have on the amount of hardware infrastructure that would be required to support financial transactions are immense. The number of server and storage systems needed for back-office applications in theory would drop substantially, as would the amount of power we need to support that infrastructure. Essentially, it would change the underlying economics of enterprise computing. It would also change the dynamics of application development because we would no longer be relying on hundreds of thousands of lines of code to process transactions.

We're still in the very early days of parallelization, so it's too early to say with absolute certainty that N_gine is on to something. But Hinkle's ideas on a new approach to back-office applications are starting to get hearings in the corridors of Wall Street and other IT organizations that are looking for new approaches to building and running financial transaction systems at a cost that are several magnitudes less than what we're spending today.