A 3-Level Cache Miss Model for a Nonvolatile Extension to Transcendent Memory

Abstract

Resource allocation is fundamental to cloud computing, where the memory hierarchy is deep. Space allocation in this hierarchy calls for a model to determine how provisioning at one level affects performance at a lower level. This paper presents a 3-level model that relates the Miss Ratio Curves for two caches at adjacent levels. The model is tested with NEXTmem, which is a transcendent memory used by a Xen hypervisor to cache pages for virtual machines. NEXTmem has a DRAM level and a nonvolatile memory level. The test runs DaCapo benchmarks and shows that the model can be used to enforce fairness at one level, and latency bounds at another level.