Scaling Switch-driven Flow Control with Aquarius

Abstract

As datacenter networks support more diverse applications and faster link speeds, effective end-to-end congestion control becomes increasingly challenging due to the inherent feedback delay. To address this issue, switch-driven per-hop flow control (FC) has gained popularity due to its natural flow isolation, timely control loop, and ability to handle transient congestion. However, the ideal FC requires impractical hardware resources, and the state-of-the-art approximation approach still demands a large number of queues that exceeds common switch capabilities, limiting scalability in practice. In this paper, we propose Aquarius, a scalable solution for per-hop FC that maintains satisfactory flow isolation with a practical number of queues. The key idea of Aquarius is to take independent control of different flows within the same queue, discarding the traditional practice of managing traffic collectively within the same queue. At its core, Aquarius applies a contribution-aware pausing mechanism on congested switches to enable individual control decisions for arriving flows, and uses an opportunistic re-assigning strategy on upstream switches to further isolate congested and victim flows. Experimental results demonstrate that Aquarius maintains comparable performance with 4 × fewer queues, and achieves 5.5 × lower flow completion times using the same number of queues, compared to existing solutions.