CAREER: SentientCache: Rethinking the Cache Abstraction

Emory University, Atlanta GA

Investigators

Abstract

The SentientCache project focuses on improving the performance and cost-effectiveness of large-scale distributed systems by automatically learning what data should be memorized. Inadequate performance of distributed systems, including those supporting major websites, is frustrating and costly. Performance problems are frequently addressed by using dedicated computers, so-called cache servers, for storing answers to common requests to reduce response time while also shielding the databases supporting the system from disruptively high loads. The SentientCache project aims to revamp the design of modern caches to better utilize server resources. The project will produce theoretically and experimentally validated prototypes of cache systems that factor in external costs and contextual information, and learn from experience. The SentientCache project will engage graduate and undergraduate students in innovative research through real-world workload analysis, software engineering and large-scale system experimentation. As well, there is an educational outreach plan to expose undergraduates, high school and K-12 students to research in computer systems with exciting competition components for high-school and undergraduate students to promote computer science careers. Successful research outcomes from the project should advance the field of computer systems and shift the cache paradigm towards endowing system caches with greater intelligence and interaction with their environments in several ways: The research will systematically characterize changes in workloads, scale, performance requirements and memory constraints of distributed caches, use advanced machine learning and other prediction techniques within systems research, and better determine how profiling can help to automatically perform cross-cutting performance optimization within large-scale systems.