PSP: Low-Overhead Packet-Level Load Balancing for Stale-State and Bandwidth-Asymmetric Networks

2026-08-08

AI Summary

Traditional flow-level load balancing often suffers from hash collisions and tail latency due to its lack of link-state awareness, while existing packet-level approaches are hindered by stale state information, high hardware overhead, and poor adaptability to heterogeneous links. This work proposes the Probabilistic State Proportional (PSP) scheduling algorithm, which innovatively combines discrete state modeling based on bandwidth intervals with localized probabilistic mapping to achieve low-overhead, highly stable packet-level load balancing without requiring real-time global information. Simulations demonstrate that PSP significantly outperforms Join-the-Shortest-Queue (JSQ) and Random across diverse network scales and interference scenarios, achieving lower packet loss rates and 99th-percentile buffer occupancy, better scalability, and performance comparable to Top-k at substantially reduced hardware cost.

Abstract

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.

Problem

• load balancing
• stale state
• bandwidth asymmetry
• packet-level scheduling
• data center networks

Innovation

• packet-level load balancing
• stale state
• bandwidth asymmetry
• probabilistic dispatching
• hardware-efficient