PREDICTIVE RESOURCE MANAGEMENT BY REDUCING COLD START IN SERVERLESS CLOUD

Abstract

Abstract Serverless Cloud computing expanding its domain rapidly. This is simple, efficient, light-weight, secure and ubiquitous. All Cloud players provide it with different attractive names such as Amazone branding it with AWS Lambda, Goole using Cloud Run, Ali Baba calling it Function Compute and last but not least Microsoft providing serverless cloud with name of Azure Function. Normally, function service executes the core business logic of application and host's machine policy of execution create a significant impact of overall quality of service provided by CSP (Cloud Service Provider). To produce an effective execution policy, the host machine maintains a lean balance between Cold and Hot restart. Policy efforts to reduce Cold restart but manage resources during Hot restart. In this paper, we employed a machine learning based classification methodology that segregate the functions in terms of cold and hot functions. We implemented Naïve Bayes classifier and boosting the accuracy with Kernel Density Estimation. The overall best accuracy was observed up to 94.35%.