Design and Implementation of Data-Intensive Application using Memory Expansion Device

Modern intelligent applications utilizing data-driven analysis have gained significant attention. As these applications rely on analyzing large-scale data for improved performance, they demand substantial high-performance memory. However, current computer architectures often lack the capacity to provide enough fast memory, leading to performance degradation in such applications. To address these limitations, memory expansion devices are being developed that connect to host servers through the new cache coherent interconnects like CXL. These devices offer additional memory and hardware acceleration capabilities to enhance performance.In this paper, we propose a novel method to accelerate data-intensive MPI applications in a CPU-FPGA heterogeneous computing environment, employing memory expansion devices called MEX. Our approach leverages MEX and MPI to enable near-memory processing by utilizing MEX as shared computing resources. To the best of our knowledge, this is the first study to combine memory expansion devices and MPI for accelerating data-intensive MPI applications. We implement a content-based image similarity search system using MPI and MEX and verify the feasibility of our proposed methods.