Achieving high utilization of flowlet-based load balancing in data center networks

Abstract Modern data center networks provide multiple paths between any host pairs. Load balancing traffic across these paths is critical to achieve low latency and high throughput. Despite prior solutions show that flowlet-based solutions are powerful in achieving load balancing in asymmetric topology, they suffer from the congestion mismatch problem in rerouting flowlets across different paths. That is, due to lacking the explicit congestion feedback, existing flowlet-based solutions are difficult to utilize bandwidth resource efficiently, which may result in low link utilization and packet loss. In this paper, we propose a congestion-aware load balancing scheme named CAF to eliminate the problem of congestion mismatch. The basic idea behind CAF is that whenever the sender perceives flowlet switching, it proactively sends probe packets to measure available bandwidth and uses the measurement result to properly set the congestion window, avoiding the unnecessary under-utilization and packet loss. Through a series of large-scale NS2 simulations and testbed experiments, we demonstrate that CAF reduces average flow completion time by up to 86% compared with the state-of-the-art mechanisms.