CHOPT: Rethinking Memory Placement Decisions

Lei Zhang, Reza Karimi, Irfan Ahmad, Ymir Vigfusson

New memory technologies, such as NVM, blur the traditionally distinctive performance characteristics of adjacent layers in memory hierarchies. Conventional assumptions of orders of magnitudes of difference between layers in request latency and space capacity no longer hold. We now face data placement challenges: which data should be cached in faster memory if it could instead be served directly from slower memory?

This project proposes an optimal offline algorithm to analyze the potential speed-up when placement decisions - whether a requested block should be brought into cache or not - can be made, even when incorporating asymmetric read and write costs. The goal is to identify significant opportunities for algorithm design to improve the data placement decisions of online memory management systems.


This work is under review.


Irfan Ahmad



National Science Foundation

Emory University