Learning to Schedule Multi-Server Jobs With Fluctuated Processing Speeds

Article Computer Science, Information Systems

DPoS: Decentralized, Privacy-Preserving, and Low-Complexity Online Slicing for Multi-Tenant Networks

Hailiang Zhao et al.

Summary: This paper proposes an online algorithm DPoS for multi-resource slicing, which is fully decentralized and low-complexity. The algorithm achieves close-to-offline-optimal performance in most cases and has low algorithmic overheads.

IEEE TRANSACTIONS ON MOBILE COMPUTING (2022)

添加到收藏夹

Article Computer Science, Information Systems

Distributed Redundant Placement for Microservice-based Applications at the Edge

Hailiang Zhao et al.

Summary: Multi-access edge computing (MEC) is a promising paradigm that pushes computation and communication resources to the network edge. We propose a distributed redundant placement framework and a GA-based Server Selection (GASS) algorithm for microservice-based applications. Numerical results based on real-world dataset confirm the superiority of our proposed approach over benchmark policies.

IEEE TRANSACTIONS ON SERVICES COMPUTING (2022)

添加到收藏夹

Article Automation & Control Systems

Evolving Scheduling Heuristics via Genetic Programming With Feature Selection in Dynamic Flexible Job-Shop Scheduling

Fangfang Zhang et al.

Summary: A novel two-stage GPHH framework with feature selection is designed in this article to automatically evolve scheduling heuristics in DFJSS, and individual adaptation strategies are proposed to utilize information. Results show that the proposed algorithm can successfully achieve more interpretable scheduling heuristics with fewer unique features and smaller sizes, and reach comparable scheduling heuristic quality with much shorter training time than traditional algorithms.

IEEE TRANSACTIONS ON CYBERNETICS (2021)

添加到收藏夹

Article Computer Science, Theory & Methods

DL2: A Deep Learning-Driven Scheduler for Deep Learning Clusters

Yanghua Peng et al.

Summary: In this article, a DL-driven scheduler called DL2 is proposed for DL clusters, which adopts a joint supervised learning and reinforcement learning approach to dynamically adjust resource allocation. The implementation of DL2 on Kubernetes demonstrates superior performance compared to traditional schedulers in terms of average job completion time.

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS (2021)

添加到收藏夹

Proceedings Paper Computer Science, Hardware & Architecture

Tailored Learning-Based Scheduling for Kubernetes-Oriented Edge-Cloud System

Yiwen Han et al.

Summary: The paper introduces KaiS, a learning-based scheduling framework for edge-cloud systems aimed at improving request processing throughput rate. Utilizing a coordinated multi-agent Actor-Critic algorithm and graph neural networks to embed system state information, the framework successfully learns appropriate scheduling policies to enhance system throughput rate.

IEEE CONFERENCE ON COMPUTER COMMUNICATIONS (IEEE INFOCOM 2021) (2021)

添加到收藏夹

Article Computer Science, Theory & Methods