Benchmarks that Model Enterprise Workloads
In this paper we demonstrate the use of macrobenchmarks in Linux kernel development. We describe two macrobenchmarks, SPECjAppServer2002 benchmark application and IBM's Trade, which are based on the Java platform and modeling enterprise applications typically found in large data centers. This paper shows how these macrobenchmarks were used to analyse potential improvements in the load balancing and yield behaviour of the 2.5 kernel's O(1) CPU scheduler. We also demonstrate how the macrobenchmarks helped debug the 2.5 kernels and compare their performance improvements over the 2.4 series.