23 March 2026 in IEEE Transactions on Networking
Training Mixture-of-Experts (MoE) models introduces sparse and highly imbalanced all-to-all communication that dominates iteration time. Conventional load-balancing methods fail to exploit the deterministic topology of Rail architectures, leaving multi-NIC bandwidth underutilized. We present RailS, a distributed load-balancing framework that minimizes all-to-all completion time in MoE training. RailS leverages the Rail topology’s symmetry to prove that uniform sending ensures uniform receiving, transforming global coordination into local scheduling. Each node independently executes a Longest Processing Time First (LPT) spraying scheduler to proactively balance traffic using local information. RailS activates N parallel rails for fine-grained, topology-aware multipath transmission. Across synthetic and real-world MoE workloads, RailS improves bus bandwidth by 20%–78% and reduces completion time by 17%–78%. For Mixtral workloads, it shortens iteration time by 18%–40% and achieves near-optimal load balance, fully exploiting architectural parallelism in distributed training.
10 February 2026 in IEEE Transactions on Network and Service Management
Community GPU(Graphics Processing Unit) platforms are emerging as a cost-effective and democratized alternative to centralized GPU clusters for AI(Artificial Intelligence) workloads, aggregating idle consumer GPUs from globally distributed and heterogeneous environments. However, their extreme hardware/software diversity, volatile availability, and variable network conditions render traditional schedulers ineffective, leading to suboptimal task completion. In this work, we present REACH (Reinforcement Learning for Efficient Allocation in Community and Heterogeneous Networks), a Transformer-based reinforcement learning framework that redefines task scheduling as a sequence scoring problem to balance performance, reliability, cost, and network efficiency. By modeling both global GPU states and task requirements, REACH learns to adaptively co-locate computation with data, prioritize critical jobs, and mitigate the impact of unreliable resources. Extensive simulation results show that REACH improves task completion rates by up to 17%, more than doubles the success rate for high-priority tasks, and reduces bandwidth penalties by over 80% compared to state-of-the-art baselines. Stress tests further demonstrate its robustness to GPU churn and network congestion, while scalability experiments confirm its effectiveness in large-scale, high-contention scenarios.
20 May 2026 in IEEE Transactions on Network Science and Engineering
Distributed deep neural network (DNN) training in multi-tenant cloud environments generates long-lived, bandwidth-intensive, and interdependent flows that challenge traditional fairness-based network scheduling. We present Gsched, a framework that orchestrates centralized planning with hardware priority enforcement to realize communication interleaving, thereby eliminating contention and maximizing pipeline efficiency. Unlike methods relying on complex rate control or custom switch hardware, Gsched leverages standard commodity switch features like DSCP and SPQ to enforce strict priority scheduling. It employs a cold-start profiling mechanism to capture task characteristics without user intervention, and a centralized solver to compute the optimal interleaving order that maximizes Average Transmission Efficiency (ATE). Our analysis proves that this prioritization strategy deterministically drives the system to a contention-free state, while a local perturbation theorem guarantees robustness against distributed execution conflicts via network backpressure. Across workloads, Gsched achieves performance close to the state-of-the-art Crux under heterogeneous workloads and remains competitive in highly diverse cases. More importantly, Gsched provides a theoretical foundation that justifies scalable per-bottleneck scheduling in multi-bottleneck networks.
11 June 2026 in IEEE Transactions on Cloud Computing
The rapid growth of large language model (LLM) services imposes increasing demands on distributed GPU inference infrastructure. Most existing scheduling systems follow a reactive paradigm, relying solely on the current system state to make decisions, without considering how task demand and resource availability evolve over time. This lack of temporal awareness in reactive approaches leads to inefficient GPU utilization, high task migration overhead, and poor system responsiveness under dynamic workloads. In this work, we identify the fundamental limitations of these instantaneous-state-only scheduling approaches and propose Temporal Optimal Resource scheduling via Two-layer Architecture (TORTA). TORTA introduces a spatiotemporal scheduling framework that captures both long-term workload patterns and short-term execution constraints. It adopts a two-layer design: a macro-level scheduler leverages reinforcement learning and optimal transport to coordinate inter-region task distribution, while a micro-level allocator refines task-to-server assignments within each region to reduce latency and switching costs. Experimental results across multiple network topologies show that TORTA reduces average inference response time by up to 15%, improves load balance by approximately 4-5%, and cuts total operational cost by 10-20% compared to state-of-the-art baseline methods.
19 August 2026 in Journal of Network and Computer Applications