Remote direct memory access (RDMA) has generated considerable interest in applying it in modern datacenters. Nevertheless, to fully utilize its high performance, we need designs to bridge the semantic gap between systems and RDMA's hardware features and optimizations guidelines to coordinate it with other hardware technologies (e.g., NVM).
In this talk, I will present our recent efforts in building high-performance systems with RDMA. First, I will introduce XStore, a network-attached key-value store that uses a machine learning approach to reduce the RDMA operations required for index traversal from O(log N) to O(1). Next, I will show our systematic study to efficiently coordinate RDMA and non-volatile memory (NVM) together. Finally, I will present how we provides RDMA-based primitives to accelerate serverless computing applications.
Please email for a
Xingda Wei is an assistant professor at the Institute of Parallel and Distributed Systems, Shanghai Jiaotong University. He received a Ph.D. degree (Advisors: Prof. Binyu Zang, Prof. Rong Chen and Prof. Haibo Chen) from the university. His research focuses on improving the performance and reliability of parallel and distributed systems, including co-designing systems with advanced heterogeneous hardware technologies (e.g., RDMA, NVM and SmartNIC), machine learning for systems and systems for next-generation cloud computing infrastructure. He has published papers on top system venues, including SOSP and OSDI and has won a Microsoft Research Asia Ph.D. Fellowship award in 2018.