Investigate and validate elapsed-time-based hybrid simulation extension

Investigate and validate the proposed extension to Schwetman’s hybrid simulation method that adjusts each task’s residual service demand according to elapsed time rather than counting remaining processing cycles, specifically by updating a task i’s device-specific service demand via X_d^i := X_d^i (1 − t/R_i), where t is the elapsed time and R_i is the mean residence time obtained from a closed product-form queueing network solution, to assess its suitability for performance evaluation of multiprogrammed computer systems and task systems.

Background

The paper reviews hierarchical performance analysis techniques for queueing-network-based models and highlights Schwetman’s hybrid simulation as a flexible higher-level approach. The traditional hybrid simulation method tracks remaining processing cycles per task and recomputes cycle times as concurrency changes.

The authors propose a generalization that does not require cyclic processing: instead of counting remaining cycles, each task’s residual service demands are reduced in proportion to the elapsed time relative to the task’s mean residence time obtained from the lower-level queueing network model. This modification is intended to simplify implementation and potentially improve accuracy when completion times are not well-captured by exponential assumptions, but the authors explicitly note that further investigation and validation are necessary.