Near-Data Processing in CHORYS

Abstract Data-intensive applications in the cloud are expected to service an ever-increasing number of users and handle exponentially growing volumes of data. Meeting these expectations is a challenge because they require balanced performance across the full memory hierarchy, from processor caches to disks accessed over the network. The problem is compounded by the cloud’s shift toward disaggregated architectures, where memory and storage are pooled across fabrics, thus lengthening the data path. CHORYS addresses this challenge through near-data processing with open, programmable accelerators embedded in storage and network devices. CHORYS considers an accelerator-based cloud model in which Solid State Drives (SSDs) and Network Interface Cards (NICs) execute data-intensive operations such as filtering, aggregation, and parsing directly on the data path, thus reducing latency, energy consumption, and resource usage. The project contributes both hardware platforms (OpenSSD‑V and OpenNIC‑V, built around RISC‑V cores) and a software stack including data-intensive pattern libraries, APIs, and runtime systems for the seamless integration of near-data execution into existing cloud infrastructures. CHORYS use cases illustrate near-data processing in the context of data lakes, object stores, and AI inference pipelines.