Transforming an Interactive Expert Code into a Statefull Service and a Multicore-Enabled System

Legacy codes are valuable assets that are difficult or even impossible to be rebuild each time when the underlying computing architecture is changed at conceptual or physical levels. New software engineering technologies, like the ones supporting the concept of service oriented architecture, promises to allow the easy use of the legacy codes. Despite this promise, the transition towards a service oriented architecture is not a straightforward task especially for the legacy codes with a rich user interface: the part mostly affected by the architectural change is the interface. This paper describes the transforming path that was followed in the case of a ten years old expert system for solving initial value problems for ordinary differential equations, emphasizing how the new interface should look like to preserve the code functionality despite the conceptual switch from a human user to a software code. The use of statefull Web services allows also preserving the large database of problems and methods of the expert code, as well as its special functionality that allows the extension of this database, in the benefit of any user of the service, human or another software code. This transformation path can be easily followed by other similar legacy codes, especially those designed for scientific computing. Moreover, the part of the expert system specially designed to deal with parallel computing techniques is extended in order to take advantage of the latest achievements in what concerns the hardware architectures, more precisely, to be able to exploit the advantages of multicore systems. In this context it is proved that parallelism across the numerical method that requires a small number of processors and can profit from the shared memory available to all cores can improve the response time of the expert code.