Exploring Allocation Policies in Disaggregated Non-Volatile Memories

Many modern applications have memory footprints that are increasingly large, driving system memory capacities higher and higher. However, due to the diversity of applications that run on High-Performance Computing (HPC) systems, the memory utilization can fluctuate widely from one application to another, which results in underutilization issues when there are many jobs with small memory footprints. Since memory chips are collocated with the compute nodes, this necessitates the need for message passing APIs to be able to share information between nodes. To address some of these issues, vendors are exploring disaggregated memory-centric systems. In this type of organization, there are discrete nodes, reserved solely for memory, which are shared across many compute nodes. Due to their capacity, low-power, and non-volatility, Non-Volatile Memories (NVMs) are ideal candidates for these memory nodes. Moreover, larger memory capacities open the door to different programming models (more shared memory style approaches) which are now being added to the C++ and Fortran language specifications. This paper proposes a simulation model for studying disaggregated memory architectures using a publicly available simulator, SST Simulator, and investigates various memory allocation policies.