Search papers, labs, and topics across Lattice.
This paper introduces the probabilistic state-proportional (PSP) dispatching algorithm for packet-level load balancing, addressing the limitations of traditional hash-based methods in data center networks. By utilizing a Band-based discrete state representation, PSP minimizes hardware complexity and effectively manages stale state information, leading to improved performance in multipath Clos networks. Experimental results demonstrate that PSP significantly reduces loss rates and buffer occupancy while maintaining scalability, outperforming existing methods like join-the-shortest-queue and Random scheduling.
PSP achieves superior load balancing in AI data centers by effectively managing stale state information while minimizing hardware complexity.
With the rapid growth of large language model training and generative artificial intelligence services, data center networks face severe micro-burst traffic and high concurrency. Traditional hash-based flow-level load balancing cannot sense link states, leading to hash collisions, hotspot congestion, and tail latency in multipath Clos networks. Existing packet-level schemes are constrained by stale state information, high hardware complexity, and poor adaptation to heterogeneous links. To address these issues, this paper proposes probabilistic state-proportional (PSP) dispatching, a packet-level load balancing algorithm. Using a Band-based discrete state representation, PSP replaces global sorting with local probability mapping, reducing hardware complexity while suppressing herding and oscillations caused by stale states. Experiments on a cycle-accurate simulator show that PSP is robust across port scales, bandwidth-limited paths, and fixed-flow interference. It outperforms join-the-shortest-queue (JSQ) scheduling and Random in loss rate, 99th-percentile buffer occupancy, and scalability, while remaining competitive with Top-k at lower hardware cost. PSP provides an effective balance among performance, stability, and overhead for artificial intelligence data centers.
