CSR: Small: Squeezing More Performance Out of Distributed Storage Systems With a Transparent Ordering-Control Layer

Distributed storage systems are a fundamental enabler of today’s web applications, which are ubiquitous in people’s daily lives and have an unparalleled impact on society. In principle, academic research should be able to directly benefit these systems, playing a vital role in improving social welfare. However, existing research innovations rarely reach production. This proposal designs novel techniques to improve the performance of deployed systems without having to change their codebase, and thus ensures ease of adoption which leads to direct practical impacts. The proposed research will advance the state of the art in the design of distributed storage systems and help with the translation of academic innovations into practical deployments. The project will build STEM talent and broaden participation by training both undergraduate and graduate students, providing research opportunities for underrepresented groups, and inspiring curriculum design that integrates teaching and research. This project aims to understand the impact of request ordering on the performance of concurrency control and replication, which are two critical components of distributed storage systems, and design solutions that improve the system's performance without having to change its currently deployed concurrency control and replication techniques. The scope of this project includes distributed transactions, consistency models, concurrency control, and replication. The project will take an experiment-driven approach to understand the implication of request ordering on system performance and develop new designs by leveraging request ordering. The project will evaluate the solutions at scale against state-of-the-art techniques and make their details publicly available. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.