The IRMA Community
Newsletters
Research IRM
Click a keyword to search titles using our InfoSci-OnDemand powered search:
|
Adaptation of a Parallel Processing Technique Used to Solve a Physics Problem to a Computer Network Management Application
Abstract
Parallelization of numeric calculations and development of new parallel algorithms is an important aspect for successful computational investigations in many fields of science. Recently the advantages of parallel processing have been recognized as applicable in the business world as well [1]. For many computational problems a single processor can easily provide timely results. However, there are certain classes of problems in which a single processor alone is not adequate. It may be that the computations or the I-O are so intensive that the single processor will need help to complete the functions quickly enough to be of value to the end-user [2]. If that is the case, parallel processing resources are more readily available and effective than they were even five years ago [3]. The combination of high speed networks such as 100BASETX, high end workstations, clustering software and middleware protocols such as MPI (message passing interface) allow a group of high end workstations connected by a LAN and configured with the appropriate software to become a loosely-coupled scalable parallel computing architecture [4]. In the past, some of the hesitation to employ parallel processing was related to its high cost. Configuring existing equipment to serve in a dual role negates much of that argument. Therefore, if an appropriate software candidate for parallel processing is identified the next concern is often related to software development. In other words, how is that software going to be effectively broken into subparts? Past work has shown three outcomes when implementing parallel processing algorithms in regard to speed: parallel is faster than serial, serial is faster than parallel, there is no significant difference between serial and parallel [5]. Obviously if the first outcome is not realized parallel processing is not worth pursuing. Developing parallel processing algorithms can be a time consuming and frustrating task and often there is no guarantee that it will speed up processing. Therefore, when trying to solve a complex problem that is a candidate for parallel processing it is often beneficial to investigate if similar algorithms exist. If so it may be efficient to try to adapt a proven algorithm rather than devise one from scratch. Therefore, this paper is designed to serve as a case study in which a proven scientific algorithm is adapted to a management problem. Specifically, the case selected is a parallel plotting algorithm used to solve a semiquantal few-body physics problem, which in turn is modified to solve a packet inter-arrival time distribution problem in the network management domain.
|
|