Remote Direct Memory Access (RDMA) is a widely adopted optimization strategy in datacenter networking that surpasses traditional kernel-based TCP/IP networking through mechanisms such as kernel bypass and hardware offloading. However, RDMA also faces a scalability challenge with regard to connection management due to limited on-chip memory capacity in the RDMA Network Interface Card (RNIC). This necessitates the storage of connection context within RNIC’s memory and induces considerable performance degradation when maintaining a large number of connections. In this paper, we propose a novel RNIC microarchitecture design that achieves peak performance and scales well with the number of connections. First, we model RNIC and identify two key factors that degrade performance when the number of connections grows large: head-of-line blocking when accessing the connection context and connection context dependency in transmission processing. To address the head-of-line blocking problem, we then combine a non-blocking connection requester and connection context management module to process prepared connections first, which achieves peak message rate when the number of connections grows large. Besides, to eliminate connection context dependency in RNIC, we deploy a latency-hiding connection context scheduling strategy, maintaining low latency when the number of connections increases. We implement and evaluate our design, demonstrating its successful maintenance of peak message rate (66.4 Mop/s) and low latency (3.89 µs) while scaling to over 50,000 connections with less on-chip memory footprint.

远程直接内存访问 (RDMA) 是数据中心网络中广泛采用的优化策略，它通过内核旁路和硬件卸载等机制超越了传统的基于内核的 TCP/IP 网络。然而，由于 RDMA 网络接口卡 (RNIC) 的片上内存容量有限，RDMA 也面临着连接管理方面的可扩展性挑战。这需要在 RNIC 内存中存储连接上下文，并且在维护大量连接时会导致相当大的性能下降。在本文中，我们提出了一种新颖的 RNIC 微架构设计，该设计可实现峰值性能并可根据连接数量进行良好扩展。首先，我们对 RNIC 进行建模，并确定当连接数量变大时会降低性能的两个关键因素：访问连接上下文时的队头阻塞和传输处理中的连接上下文依赖性。为了解决队头阻塞问题，我们将非阻塞连接请求者和连接上下文管理模块结合起来，首先处理准备好的连接，这在连接数量变大时实现了峰值消息速率。此外，为了消除RNIC中的连接上下文依赖，我们部署了延迟隐藏连接上下文调度策略，在连接数量增加时保持低延迟。我们实施并评估了我们的设计，证明其成功维持了峰值消息速率（66.4 Mop/s）和低延迟（3.89 µs），同时以更少的片上内存占用扩展至超过 50,000 个连接。