The Esprit-funded Dynamite project team in two years' time has developed a dynamic execution programme which handles automatic work-load balancing of Parallel Virtual Machine (PVM) tasks in workstation cluster environments with dynamically changing load to applications by migrating individual tasks in a manner which is robust, efficient and transparent to the user and the application programmer. Dynamite monitors the load on the workstation cluster and, if necessary, dynamically redistributes tasks again to more efficiently use the hardware available.
The marketing study on the usability of Dynamite in health care applications has tried to assess the various medical scenarios in which the software may be of significant use to save precious time in high performing computational calculation jobs. To start with, Dynamite requires large applications, able to run on a cluster of workstations, and preferably a mixed environment with a number of users or applications which run simultaneously, or an application which in itself is highly unbalanced. The medical applications that are most eligible for Dynamite implementation are simulation and image processing, since both of them involve an enormous amount of computational power.
However, observation and evaluation of the medical market has learned that the world of parallel processing in both medical domains of simulation and 3D visualisation is far from uniform because many of these systems are still under development. The design of sophisticated algorithms to reconstruct 3D images from magnetic resonance imaging (MRI), computed tomography (CT) or positron emitter tomography (PET) scans requires a lot of effort and research. Very much the same goes for virtual reality applications, such as haptic feedback and graphical interfaces. In these areas, the computational power is delivered by workstations as well as by supercomputers.
As a more suitable field for Dynamite implementation, the marketing study indicates the discipline of image volume rendering. Preferably, a message passing interface (MPI) version of the software should be developed in order to create a competitive advantage. The renowned hardware vendors however are not very inclined to act as co-designers since this would ruin their own server market. The area of operative planning systems looks promising as well for Dynamite, rather than the training simulators and operative support systems, which require a real time component, and specific hardware as an indispensable consequence, to obtain a satisfactory and realistic functioning result.
The marketing report also states that in the past recent years, many projects funded by the European Commission have actively focused on the subject of medical parallel processing. Some of the older projects, based on PVM, have used Parsytec hardware to run their programmes. These are of a particular interest to Dynamite for immediate implementation. The team could equally envisage to collaborate with Linux companies, which form an alternative for the as yet dominant position of the Windows NT platforms.
The study concludes that in addition to the valuable option of designing an MPI version of the Dynamite software, it might still be more appropriate to fully concentrate on the market of Mechanical Computer Aided Engineering (MCAE), since here parallel processing is more frequently used than in other fields. For more technical details on the project, please consult the Dynamite Web site.