| 摘 要: 针对企业云原生环境中Kubernetes默认调度器在异构资源和多样化工作负载调度中暴露出的响应延迟高、资源碎片化严重和集群稳定性不足等行业共性问题，展开优化研究。尽管已有研究采用多目标决策和机器学习等方法对调度策略进行改进，但仍普遍存在优化目标单一、动态适应能力弱、对实时敏感任务支持有限以及对集群健康状态关注不足等缺陷。为此，本文提出一种任务驱动的Kubernetes异构任务调度模型，该模型结合改进的TOPSIS（逼近理想解）多准则决策与事件触发机制，实现对多维资源的动态评估与节点综合排序，并通过资源预留及并行绑定策略显著降低调度延迟。实验表明，本模型在延迟、IO及内存压力等多个性能指标上均显著优于传统调度器及典型优化算法，能够有效提升高并发异构环境下的资源利用率与系统稳定性，为行业提供了一种兼顾效率、实时与稳定性的调度解决方案。 |
| 关键词: 任务调度 Kubernetes 逼近理想解法 任务驱动 延迟优化 集群健康 内存优化 |
|
中图分类号:
文献标识码:
|
| 基金项目: 国家自然科学基金（No.92067201）； 江苏省重点研发计划（No.BE2020084-4） |
|
| A TASK-DRIVEN HETEROGENEOUS TASK SCHEDULING MODEL BASED ON KUBERNETES |
|
zhaoyi
|
nanjing university of posts and telecommunications
|
| Abstract: In response to the common industry issues exposed by Kubernetes' default scheduler in heterogeneous resource and diverse workload scheduling—such as high response latency, severe resource fragmentation, and insufficient cluster stability—this paper conducts optimization research. Although existing studies have improved scheduling strategies using multi-objective decision-making and machine learning, they still generally suffer from shortcomings including single optimization objectives, weak dynamic adaptability, limited support for real-time sensitive tasks, and insufficient attention to cluster health status.To address these limitations, this paper proposes a task-driven Kubernetes heterogeneous task scheduling model that combines an improved TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) multi-criteria decision-making approach with an event-triggered mechanism. This enables dynamic evaluation of multi-dimensional resources and comprehensive node ranking, while significantly reducing scheduling latency through resource reservation and parallel binding strategies.Experimental results demonstrate that the proposed model outperforms traditional schedulers and typical optimization algorithms across multiple performance metrics including latency, I/O, and memory pressure. It effectively improves resource utilization and system stability in high-concurrency heterogeneous environments, providing an industry solution that balances efficiency, real-time performance, and stability. |
| Keywords: Task Scheduling Kubernetes Approaching the Ideal Solution Task-Driven Latency Optimization Cluster Health Memory Optimization |
