Collaborative Research: CSR: Reinventing the System Stack for Automated Compute Scaling with Compute Express Link (CXL) Shared Memory

Cloud computing lets people run programs on powerful computers in data centers. To handle bigger jobs, these programs often need to spread their work across many machines at the same time. But most programs were built to run on just one computer, and modifying them to cooperate across many computers is slow, difficult, and prone to mistakes. This project explores software support for a new generation of computer hardware that lets separate machines share the same memory, almost as if they formed one large computer. The goal is to help ordinary programs run faster and at a much larger scale without having to be rewritten from scratch. The intellectual merit centers on a complete system stack that spans programming models, compilers, runtime systems, and operating systems, all built around Compute Express Link (CXL), a hardware standard that allows multiple machines to share memory with cache coherence. The first task delivers fine-grained sharing of individual memory locations together with safe synchronization abstractions. The second task creates compiler and runtime support that migrates program objects dynamically across different memory regions. The third task adds transparent page-based memory sharing inside the operating system so that unmodified applications scale automatically. The fourth task builds a dynamic analysis tool that detects coherence errors in software that relies on software-coherent memory. The broader impact lies in making cloud computing more affordable and more efficient. By enabling existing single-machine programs to run across many servers without rewriting, the project can improve server utilization, reduce hardware overprovisioning, and lower the operating cost of data centers. It especially benefits research laboratories, academic institutions, and small companies that lack large infrastructure but depend on scalable computing for artificial intelligence, data analytics, and scientific discovery. The work enriches courses in parallel computing, operating systems, and cloud systems, and it recruits graduate, undergraduate, and high school students into research. 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.