Analysis of Memory Sensitive SPEC CPU2006 Integer Benchmarks for Big Data Benchmarking

Abstract

Benchmarking for Big Data is done at the system level, but with processors now being designed specifically for Cloud Computing and Big Data applications, optimization can now be done at the node level. The purpose of this work is to analyze three SPEC CPU2006 Integer benchmarks (libquantum, h264ref and hmmer) that were deemed "highly memory sensitive" in other works to determine their potential as Big Data processor benchmarks. Program characteristics like instruction count, instruction mix, locality, and memory footprint were analyzed. Through this preliminary analysis, these benchmarks were determined to be potential Big Data node-level benchmarks, but more analysis will have to be done in future work.