CSR: Small: Distributed Shared Persistent Memory
Purdue University, West Lafayette IN
Investigators
Abstract
This project has the potential to revolutionize datacenter systems by developing the first type of system that integrates traditional datacenter memory and storage systems in one layer. The proposed software platform can potentially be used to develop various course and research projects. The investigator will publicly release the developed software to enable a rich set of datacenter applications and to foster more cross-layer research activities. Next-generation non-volatile memories (NVMs) provide byte addressability, persistence, and low-latency performance. They are poised to radically alter the landscape of memory and storage technologies and have already inspired a host of research projects. Most previous research on NVMs has focused on using them in a single machine. It is unclear how to best utilize NVMs in distributed, datacenter environments. This project takes a significant step towards the goal of using NVMs in distributed datacenter environments by developing Distributed Shared Persistent Memory (DSPM), a framework that uses a pool of machines with NVMs to form a global, shared, and persistent memory space. Applications can perform traditional memory load and store instructions to access both local and remote data in this global memory space and can at the same time make their data persistent and reliable. Unlike traditional two-layer approaches with a memory and a storage layer, DSPM has just one layer that serves both as distributed memory and as distributed storage. Anticipated deliverables include a new kernel-based, software DSPM system and a set of datacenter applications ported to DSPM.
View original record on NSF Award Search →