High-Performance Packet Routing Acceleration for Cloud Systems Using High Bandwidth Memory

IEEE Open Journal of the Communications Society

Tomohiro Ishihara¹  Hiroshi Murakawa²  Kazuki Hyoudou²  Takashi Shimizu² 

[1] ICT System Laboratory, Fujitsu Laboratories Ltd., Kawasaki, Japan;Infrastructure System Department, Fujitsu Ltd., Kawasaki, Japan;

关键词: Accelerator architecture;    field-programmable gate arrays;    high bandwidth memory;    virtual router;    hardware offload;   

DOI  :  10.1109/OJCOMS.2022.3142249

来源: DOAJ

【 摘 要 】

Virtual routers (vRouters) in cloud systems are in great demand due to the ever-increasing network bandwidth. To meet this demand, hardware offload technologies are mandatory for vRouters. Furthermore, offload hardware needs new architecture to achieve a throughput of several hundred Gbps, because memories that store routing tables will create a bottleneck. Therefore, we propose a new hardware architecture using high bandwidth memories (HBMs) and an acceleration engine equipped with multi-pipeline processing. In this paper, we improved the performance of the random memory-accesses from multiple pipelines by distributing to the memory channels of HBMs. In our prototype, vRouter packet processing performances were achieved 320 M packet per second(pps) and 250 Gbit per second (bps). Moreover, the prototype showed low and stable latency of packet transfers that was not affected by background traffic.

【 授权许可】

Unknown