文献类型：期刊文献

中文题名：基于函数关联集的冷启动优化策略研究

英文题名：Research on Cold-start Optimization Strategy Based on Function Association Set

作者：郑元伟 李子鹏 龙诺亚 张菡 张猛 宋磊 王喜宾

第一作者：郑元伟

机构：[1]贵州电网有限责任公司,贵州贵阳550002;[2]贵州大学计算机科学与技术学院,贵州贵阳550025;[3]贵州理工学院大数据学院,贵州贵阳550025

第一机构：贵州电网有限责任公司,贵州贵阳550002

年份：2025

卷号：43

期号：11

起止页码：15-21

中文期刊名：机械与电子

外文期刊名：Machinery & Electronics

基金：贵州电网有限责任公司创新项目(GZKJXM2O220069)。

语种：中文

中文关键词：Serverless;冷启动;协同预热;调度优化;FP-Growth

外文关键词：Serverless;cold start;collaborative pre-warming;scheduling optimization;FP-Growth

摘要：Serverless计算中的云函数冷启动问题会导致显著的性能开销。现有预热策略在以应用为调度单位时,常因调度粒度过粗而造成资源浪费,且难以对调用模式不清晰的函数进行有效优化。为此,提出一种基于函数关联集的冷启动优化策略FS-Warm。该策略引入“函数关联集”作为新的调度粒度,通过挖掘函数间的共现关系(如使用FP-Growth算法),将业务逻辑或调用行为上强关联的函数(包括高频函数及其关联的低频或模式不清晰函数)聚合。基于函数关联集进行协同预热与调度,旨在更精准地按需加载资源,并利用已知模式函数的调用行为辅助优化集内其他函数的预热时机,从而有效缓解冷启动问题,提高资源利用率并改善对无清晰调用模式函数的优化效果。实验结果表明,基于函数关联集的冷启动优化策略FS-Warm在Serverless云函数冷启动率的降低和内存浪费的平衡上均具有较好的表现,在75分位点上将冷启动率降至20.8%,相较于基线策略优化了42.4%,并减少了约75%的内存浪费。
The cold start problem of cloud functions in serverless computing can lead to significant performance overhead.Existing prewarming strategies,when using applications as the scheduling unit,often result in resource wastage due to overly coarse scheduling granularity and struggle to effectively optimize functions with unclear invocation patterns.This paper proposes FS-Warm,a cold start optimization strategy based on function association sets.This strategy introduces“function association sets”as a new scheduling granularity,mining co-occurrence relationships between functions(e.g.,using the FP-Growth algorithm)to aggregate functions that are strongly correlated in business logic or invocation behavior(including high-frequency functions and their associated low-frequency or pattern-unclear functions).By performing collaborative prewarming and scheduling based on function association sets,the strategy aims to load resources more precisely on demand and leverage the invocation behavior of known-pattern functions to assist in optimizing the prewarming timing of other functions within the set.This effectively mitigates the cold start problem,improves resource utilization,and enhances optimization for functions without clear invocation patterns.Experimental results show that FS-Warm,the cold start optimization strategy based on function association sets,performs well in reducing the cold start rate of serverless cloud functions and balancing memory wastage.At the 75th percentile,it reduces the cold start rate to 20.8%,a 42.4%improvement over baseline strategies,while cutting memory wastage by approximately 75%.