Parallel Computing in Water Resource Engineering
By Robert M. Wallace, Ph.D., P.E.
It seems that every few months the computer industry announces that another computer "revolution" is taking place and that we must upgrade to the latest, greatest, and most expensive technology. However, most advances in the civil engineering computing community have dealt with user input and output while the fundamental analysis programs have changed only incrementally. Because of faster individual computers, we can solve larger and more complex problems. However, we typically still use the same modeling approach that was implemented 25 years ago with the advent of personal computing. The convergence of new computing technologies and modeling approaches, however, are rapidly changing this status quo.
Parallel Computing Hardware
The expense and complexity of massively parallel super-computers have typically relegated them to the domain of a small handful of government, industrial or military research scientists. Recent advances are making this technology substantially easier to access for common computational tasks. One of the more intriguing advances in this regard is the Beowulf Cluster (http://www.beowulf.org/index.html) . These systems are simply a group of individual personal computers tied together with an inexpensive network switch. Software libraries, such as MPI, allow computational programs to use these computers as a single, distributed memory computer system. Beowulf clusters can be created from any combination of computers, ranging from high-priced engineering workstations, to low-cost personal computers. A different technology, known as collaborative computing(http://p2pjournal.com/main/collaboration.htm), uses unused computer time on many individual computers to create a pseudo-computer with very large computing capacity. Both of these existing technologies can be used to create extremely powerful computers for a mere fraction of the cost of traditional super-computers.
Parallel Computing Algorithms and Modeling Paradigms
The primary difficulty with using most parallel computers is making efficient use of the large number of available CPUs. Most analysis programs were written for sequential operation because making parallel computer software is particularly difficult and expensive. However, a new breed of analysis software and techniques is providing the basis for implementing this technology for even the most rudimentary engineering problems. Because there is a tremendous amount of uncertainty in most engineering problems, modeling approaches like Monte-Carlo and Latin-Hypercube have been employed for many years to incorporate this uncertainty directly into the solution. These types of simulations are "embarrassingly parallel" in that each Monte-Carlo simulation is independent of all other simulations. By spreading the many simulations across multiple CPUs a huge increase in efficiency can be achieved. Also, this approach can be implemented without significant modifications to the existing computational programs. Other techniques including domain decomposition and parallel matrix solvers are being incorporated into newer computational programs. These techniques allow engineers to find exact solutions for certain classes of problems by using millions of computational nodes, thus alleviating some of the stability issues inherent in complex modeling. Together, advances in parallel computing hardware and parallel analysis techniques are providing engineers and researchers with the ability to model very large, interconnected modeling scenarios such as entire river basins. Stochastic modeling can insure that uncertainty can be quickly incorporated into solutions while other parallel techniques allow for extremely detailed analyses.
This article is the fourth in a regular series of reports on emerging and innovative technologies relevant to the area of the environment and water resources produced by EWRI's Emerging and Innovative Technologies Council (EITC). EITC's mission is to advance the development, knowledge, and application of emerging and innovative technologies for the planning and management of water resources and the protection and enhancement of the environment. If you are interested in contributing an article or becoming a member of this Council, please contact Lindell Ormsbee, Chair of EITC, at lormsbee@engr.uky.edu.