Existing data-plane verifiers face severe performance challenges in verifying hyper-scale underlay data center networks (DCNs) – centralized verifiers often fail to fully exploit the parallelism offered by the modern multi-core CPUs, while distributed verifiers suffer from high overhead due to task distribution and inter-node communication. To overcome these limitations, this paper introduces HeTu, a high-performance centralized parallel data-plane verifier specifically for verifying hyper-scale underlay DCNs. HeTu achieves ultra-fast verification through three key designs: (1) a fully parallel verification framework with small graph construction overhead, (2) a new binary decision diagram management strategy that enables full parallelism by using separated storage and selectively indexing and caching network-level predicates to reduce redundant operations, and (3) an optimized forwarding graph model that aggregates parallel tasks to eliminate redundant computation. Extensive evaluations on synthetic FatTree and large-scale production datasets show that HeTu outperforms state-of-the-art algorithms in runtime by 100× to 6000×, demonstrating its superior scalability and efficiency in data-plane verification of hyper-scale DCNs.

Internet companies are constructing large-scale AI clusters with commodity Ethernet switches for AI model training to support their businesses. AI training workloads impose stringent network requirements, mandating that cluster networks deliver high peak throughput while maintaining robustness and resilience in the face of link failures. We present SGLB, a distributed, global congestion-aware load balancing system for AI clusters. SGLB operates a control-plane protocol, SyncMesh, to enable a new load balancing abstraction in modern commodity switches—Global Load Balancing (GLB) engine—which utilizes global congestion information to distribute traffic across all available paths. We address three key challenges in designing SGLB: fast routing convergence to minimize downtime in the event of link failures, scalable maintenance of congestion profiles within the constraints of limited switch hardware resources, and preventing GLB throughput suppression in scenarios where path bandwidths are asymmetric. We prototype SGLB and conduct extensive experiments to evaluate SGLB. SGLB ensures rapid routing convergence in the event of link failures, recovering in as little as 45 μs to guarantee network robustness for long-term, stable model training. Additionally, SGLB effectively load-balances traffic across paths, avoiding those with global congestion, which accelerates All-to-All collective communication by up to 60%.