SBIR-STTR Award

Technologies for parallel and distributed computation are at the forefront of software research. Vulnerability assessment studies model the interaction of a sortie - perhaps using several weapons - with a target. A single vulnerability assessment study requires the results from hundreds or thousands of independent attack scenarios won with varying input parameters. Vulnerability assessments can be parallelized at two levels. First, it is possible to distribute the individual scenarios required to perform a single study across a network of computers. Distributing scenarios creates a need for coordination and load balancing. Second, the loop that generates the various simulated interactions can be parallelized using several different approaches. Parallelizing the code internally may require a considerable rewriting of an existing code or even the development of new techniques specific to vulnerability assessment. The overall goal of this Phase I research is to develop a recommendation concerning the most beneficial approach to speeding up vulnerability assessment studies. Two approaches will be considered: distributing the scenarios of to entire study and parallelizing the interactions within a scenario. Both approaches can be implemented using a variety of techniques. During Phase I the various techniques will be evaluated and a recommendation for parallelizing vulnerability assessment studies will be developed. A limited software prototype that distributes scenarios across a network of workstations will also be implemented. The prototype will provide experience with using distributed scenarios within a study. Provided that the distributed study can be implemented with a minimum of changes to the underlying vulnerability assessment code, developing the distributed version will provide the most immediate benefits of parallelization across a variety of architectures with the least development time and cost.

Distributed Executions Everywhere (DEE) is a software tool that distributes and manages jobs (program executions) and groups of jobs across a network of Unix workstations. DEE will support both batch queue management and interactive job management while providing automatic load balancing and job distribution. Users interact with DEE using a variety of interfaces. DEE can be applied to a wide range of scientific and military applications such as parametric studies for computer models, image rendering, and vulnerability assessment studies.The performance data from DEE indicate substantial speedups without changing the EVA source code. The profile data from EVA shown in Table 1 on page 3 of the Phase I Report indicates that attempting to parallelize the outer loops of EVA would not be appropriate at this time. Discussions with other facilities indicate that a software system such as DEE would have considerable commercial potential while, at the same time, offering a high degree of support for Air Force projects.The performance data indicate that the best approach is to continue to develop DEE. Additional funding and effort will be required to move DEE into a production environment. In addition, plans for commercializing DEE are underway.

Keywords:
DISTRIBUTED COMPUTING PARALLEL COMPUTING LOAD BALANCING DISTRIBUTED OBJECTS