A Hybrid Markov Model for Accurate Memory Reference Generation

—Workload characterisation and generation is becoming an increasingly important area as hardware and application complexities continue to advance. In this paper, we introduce a concise methodology for workload generation for fast and accurate cache design space exploration. The hybrid model we propose uses an adaptation of the Least Recently Used Stack Model to capture key spatio-temporal locality features and a Markov model is implemented to generate an arbitrary length trace with the given workload characteristics through a dynamically ordered FIFO scheduler. Simulation of a variety of application traces from the SPEC2000 benchmark suite demonstrate the cacheability characteristics of the synthetic memory reference stream is generally very well preserved and similar to its original form.

• Type: Conference
• Year: 2007
• Where: IMECS
• Authors: Rahman Hassan, Antony Harris
• Keywords: Arbitrary Length Trace, IMECS 2007, Key Spatio-temporal Locality, Theoretical Computer Science, Workload Generation