Research on the CXL Memory

This paper delves into the architecture of Compute Express Link (CXL) memory technology and analyzes its current application status, challenges faced, and future trends in cloud data centers. Through experiments conducted on real systems, we evaluated the performance of CXL memory and noted that although it shows improvements in latency and bandwidth over traditional Remote Direct Memory Access (RDMA) technology, there remains a significant gap compared to local memory when handling high-concurrency and high-throughput tasks. The paper underscores the necessity of innovative memory management and scheduling strategies to bridge this gap and fully harness the potential of CXL memory. Additionally, the paper highlights the emerging industry trend towards compute-storage disaggregated architectures based on CXL memory and discusses the importance of standardization and ecosystem development for the widespread adoption of CXL technology. We anticipate that with continuous technological evolution and market maturation, optimized and innovative memory management and scheduling strategies will position CXL memory as a key player in the realm of cloud computing, offering substantial improvements in performance and efficiency.