Past Awardee

Development of Runtime Page Placement and Migration Algorithms for Merging Heterogeneous Main Memory System for Big-Data Analytic Applications

Nam Sung Kim

College: Engineering
Award year: 2016-2017

Cloud servers require a main memory system that can provide high capacity and bandwidth, as well as low latency. As the end of technology scaling is in sight, it becomes extremely challenging to increase the main memory capacity based on traditional DRAM technology.

In this project, responding to such a challenge, Kim aims to develop runtime algorithms that can predict the right memory for memory pages brought from the storage system upon page fault and decide whether or not some pages need to be migrated from DRAM to 3DX Point memory and vice versa at appropriate timing. Subsequently, Kim will integrate these developed algorithms with the Linux Virtual Memory Manager and demonstrate their efficacy using a production-grade computer system comprised of many nodes.