A Dynamic Service-Oriented Distributed Computing Framework for Evaluation of Computationally Expensive Black-Box Analyses

A practical, flexible, versatile, and heterogeneous distributed computing framework is presented that simplifies the creation of small-scale local distributed computing networks for the execution of computationally expensive black-box analyses. The framework is called the Dynamic Service-oriented Optimization Computing Framework (DSOCF), and is designed to parallelize black-box computation to speed up optimization runs. It is developed in Java and leverages the Apache River project, which is a dynamic Service-Oriented Architecture (SOA). A roulette-based real-time load balancing algorithm is implemented that supports multiple users and balances against task priorities, which is superior to the rigid pre-set wall clock limits commonly seen in grid computing. The framework accounts for constraints on resources and incorporates a credit-based system to ensure fair usage and access to computing resources. Experimental testing results are shown to demonstrate the effectiveness of the framework.