Future trends in model management systems: parallel and distributed extensions

摘要

A model management system (MMS) is a computer system which aids in the organization, utilization, and execution of models for decision makers. Currently, MMS frameworks do not provide an estimate of computer resource consumption rates for solvers, or permit alternate computational models. Without an estimate of computer resource consumption rates, the decision maker is unable to evaluate alternative solution approaches to select the most efficient. Even with such an estimate, MMS may fail the analyst if all strategies require computation time which exceeds the deadline. This is largely due to the fact the MMS only permits sequential computation. MMS prototypes limit execution of solvers to sequential, so alternative computational models such as Single Program Multiple Datastream and Multiple Input Multiple Datastream, cannot be used. MMS frameworks should be extended to provide algorithmic complexity of solvers as an evaluation measure and to incorporate parallel computational models. Networks of workstations connected by an Ethernet are an attractive low cost means of providing concurrent execution, and suit the flexible processing needs of MMS.