Jul 2026· 2026 8th International Conference on Electronics and Communication, Network and Computer Technology (ECNCT)· pp. 419-423· 0 citations· 14 references
Abstract
Adaptive queue management must balance throughput, delay, packet loss, and fairness under changing traffic and resource conditions. This paper proposes TS-D3QS, a traffic-state-aware queue scheduler that formulates multi-queue resource allocation as a discrete reinforcement-learning problem. The scheduler observes normalized queue and port features and selects one of 16 interpretable allocation profiles, each jointly specifying bandwidth shares, shared-buffer shares, and priority multipliers. A Dueling Double-DQN learner separates state value from action advantage and use a Double-DQN target to reduce value overestimation. Experiments in a reproducible four-queue simulator under bursty and non-stationary traffic show that TS-D3QS reduces latency by 7.54%, reduces aggregate loss by 3.21%, improves Jain fairness by 3.30%, and improves reward by 6.01% over vanilla DQN, while classical CoDel-like and WFQ-like rules remain competitive on selected objectives.
A hybrid reinforcement learning (RL) framework that jointly controls queue management and bandwidth allocation in bursty multi-service networks and demonstrates the effectiveness of coordinated learning-based control for stable and QoS-aware operation in bursty networked systems.
T. Khan, Babar Shah, Taimur Karamat et al.· Computing· 0 citations
Fairness between model-based TCP BBR and loss-based TCP CUBIC remains difficult when both protocols share a dynamic single bottleneck. This paper proposes OBF-SFP, a lightweight single-queue AQM mechanism that combines stochastic virtual binning, dequeue-rate shadow accounting, BDPgated activation, persistence scoring, and probabilistic penalties to reduce persistent service-share dominance. Unlike per-flow queueing, OBF-SFP estimates dominance through virtual-bin service shares while maintaining $O(1)$ per-packet processing and $O(N)$ memory cost. Evaluation in ns-3.42 uses a dynamic dumbbell topology, 2, 4, and 8 competing flows, and five random seeds. Results show that DropTail, CoDel, and PIE allow strong BBR dominance in the 8-flow case, with BBR shares around 0.889-0.892 and JFI around 0.619-0.623. OBF-SFP reduces the BBR share to 0.554 and improves JFI to 0.988, approaching FQ-CoDel fairness while retaining a single-queue design. The improvement comes with a measurable BBR-side loss trade-off, indicating that further loss reduction and ECN-capable validation remain important directions for future work.
Muhamad Arief Nugroho, Ibnu Asror, Dadang Setiawan· International Seminar on Int...· 0 citations
The rapid expansion of Large Language Models (LLMs) has established Deep Learning Training (DLT) as a major workload in modern multi-tenant GPU clusters. While fair computational scheduling is well-studied, fair communication scheduling in handling execution-time network contention remains underexplored. Current DLT communication schedulers focus primarily on efficiency-oriented objectives, often causing disparate performance degradation or even job starvation when concurrent jobs compete for shared bandwidth. In this paper, we propose LEVELLER, the first communication scheduling system that achieves max-min fairness specifically for DLT workloads. To bridge the gap between traditional flow-level fairness and DLT job-level fairness, LEVELLER introduces a novel online metric, normalized progress rate, which quantifies training experience by measuring actual progress against a contention-free ideal. We theoretically prove that a greedy, progress-rate-aware flow scheduling strategy achieves at least a 1/2-approximation of max-min fairness for arbitrary workloads, even with lazy priority updates. Building on theoretical results, LEVELLER employs preemptive, dynamic-priority scheduling to achieve near-optimal fairness on commodity infrastructure without hardware or protocol modifications. We prototype LEVELLER on both RDMA and TCP, evaluating it through a physical GPU cluster and trace-based simulations. Experimental results on 10 LLMs show that LEVELLER improves the minimum progress rate by 57% and fairness by 17% compared to state-of-the-art solutions (including CASSINI, CRUX, Symphony and Sincronia), while maintaining competitive cluster efficiency.
Geng Li, Yang Li, Mingyuan Zang et al.· Proceedings of the ACM SIGCO...· 0 citations
The demand for ultra-low latency applications necessitates advanced network architectures like the Low Latency, Low Loss, and Scalable Throughput (L4S) standard. A core component of L4S is the DualPI2 Active Queue Management (AQM), which ensures the safe coexistence of scalable and classic traffic. Despite L4S's growing adoption, the ns-3 network simulator lacks a high-fidelity, up-to-date DualPI2 model. This paper presents a comprehensive implementation of the DualPI2 AQM in ns-3, while also mirroring the official Linux Kernel architecture. Our model incorporates representative mechanisms previously absent in simulation, such as credit-based Weighted Round Robin (WRR) scheduling, step-marking, and overload protection. To guarantee simulation accuracy, we validate our implementation against a physical Linux testbed across 25 diverse Bandwidth-Delay Product (BDP) scenarios. Results demonstrate that our ns-3 model replicates real-world behavior, ensuring strict throughput fairness and queue delay isolation. Ultimately, this validated model equips the research community with a robust tool to evaluate and advance L4S performance across diverse network topologies, ranging from data centers to wireless home and office environments.
Maria Eduarda Veras, Eduardo Freitas, A. T. O. Filho et al.· 0 citations
3GPP Release 16 enables a 5G system to operate as a transparent IEEE 802.1 TSN bridge, but its scalability under heterogeneous industrial workloads remains insufficiently characterised. This paper uses the nascTime framework on OMNeT++/Simu5G to evaluate how many TSN endpoints a single 5G NR cell can bridge before per-flow QoS degrades. We model closed-loop control, machine vision, bulk telemetry, and IEEE 802.1AS traffic over a four-bearer SDAP architec- ture, varying the number of endpoints from 1 to 40, MAC scheduler, radio bandwidth (10 MHz and 20 MHz), and channel model. Results show three operating regimes. Below saturation, non-DRR schedulers perform similarly; near saturation, QoS- aware PF reduces critical-flow P99 latency by up to two or- ders of magnitude relative to channel-aware and fairness-based schedulers; and under overload, only QoS-PF maintains near- complete delivery for the highest-priority traffic. Across the two evaluated bandwidths, the saturation threshold approximately doubles when bandwidth doubles. We also show that isolating IEEE 802.1AS/gPTP traffic on a dedicated high-priority radio bearer reduces clock-servo instability, although endpoints carry- ing lower-priority data still experience elevated synchronisation delay under saturation because of reduced MAC scheduling frequency. Finally, the evaluated sub-6 GHz, 30 kHz-SCS con- figuration exhibits an effective latency floor of approximately 2.25 ms, indicating that sub-3 ms TSN deadlines may require radio-configuration changes such as configured grants or higher numerology
Mohamed A. M. Seliem, U. Roedig, C. Sreenan et al.· 0 citations
Data-center congestion control targets high throughput, fair bandwidth allocation, and low latency. Modern transports couple rate computation and packet scheduling into a single feedback loop, converging to near-optimal rates but leaving standing queues that can scale with the number of flows. We argue that separating the two reveals a simpler design point. Given stable feasible rates, the residual queue problem reduces to a timing problem: if every flow's packets arrive at the bottleneck in the correct slot, the link stays busy and the queue stays empty. Clocked ACK-Paced Synchronization CAPS is a lightweight distributed scheduling layer that achieves this by phase-locking each sender's transmissions to ACK-clocked bottleneck slots, with a per-flow correction that compensates for heterogeneous RTTs. We characterize the phase-locked steady state for dumbbell topologies under equal RTT, heterogeneous RTT, and bidirectional traffic, and validate the mechanism on a fat-tree under incast, permutation, and all-to-all traffic. CAPS reduces worst-case queue occupancy by 5-10x across all tested scenarios without throughput loss.