CAREER: Rethinking System Stack for the Load-Store I/O Era

Virginia Polytechnic Institute And State University, Blacksburg VA

Investigators

Abstract

The rapid evolution of computer hardware technology imposes challenges in bridging the gap between hardware capabilities and systems software. As data generation and processing needs grow exponentially across everyday devices like smartphones to complex systems like datacenters and supercomputers, there is a need to harness the full potential of new hardware technologies. This project focuses on Compute Express Link (CXL), a cutting-edge interconnect technology that promises to revolutionize how memory, storage, and computing devices interact within and across computers. CXL is poised to revolutionize the integration of memory and storage with near-data compute capabilities. Despite its potential, current operating systems (OS), including widely-used platforms like Linux, are not fully equipped to leverage CXL's capabilities, leading to underutilized hardware and other inefficiencies. This CAREER project endeavors to redesign the Linux kernel to optimize its compatibility with CXL technology. The overarching goal is to develop a new load-store I/O stack that can effectively manage CXL-enabled memory, storage, and compute devices. This initiative will result in enhanced performance, resource efficiency, and simplified programming requirements. The project involves a holistic approach, integrating co-design efforts across the OS, runtime environments, applications, and CXL devices. The outcomes of this research are expected to lead to significant advancements in computing, including improved memory and storage utilization, novel OS and runtime-level techniques for efficient data management, and the development of new programming models tailored for CXL-based devices. By open-sourcing all developed software, the project also promises to foster cross-disciplinary collaboration in the fields of system design and computer architecture. The project also incorporates extensive education plans to train the next-generation of computer scientists, providing students with up-to-date, research-driven learning and industry collaboration opportunities. Ultimately, this research will pave the way for more efficient, cost-effective, and high-performing computing solutions, benefiting a wide range of data-intensive applications and services, such as data-analytics, caching, key-value stores, and machine learning. 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.