Tail Latency Evaluation Under Mixed Workloads In Cloud Systems

With the development of cloud computing,the types of cloud services are increasing,and the computing demand is also increasing.The resource competition caused by multiple services running on the same platform will seriously reduce the user experience of services.In order to guarantee the performance of services,the industry mainly uses ”overoccupied” mode to ensure the response time of services.And over occupation has led to a decline in cloud data center utilization.Ensuring user experience and improving the utilization of data centers have become one of the major contradictions in cloud computing.The solution to this contradiction has been studied in many aspects in academia and industry.This paper presents and designs a benchmark test program for cloud computing system,which supports tail delay performance isolation analysis and resource utilization evaluation of cloud computing system under mixed deployment of multi-type cloud services.Firstly,the tool chooses 11 cloud services as the benchmark test load from various application areas and the degree of delay sensitivity.Secondly,the load in the benchmark test program is provided in the form of containers,which can be easily deployed on any cloud computing system with container environment.According to business scenarios,users can customize mixed different types of loads for testing;the program provides a monitoring module,which achieves real-time collection and recording of resource usage in the process of load operation,and can analyze the demand of system resources for different services through historical records;finally,the benchmark test process.In order to improve the user experience,LSTM neural network is used to predict the response latency of user requests,and the high-tail latency requests can be identified in advance.The prediction of this module allows system administrators enough time to adjust the request to other nodes for business processing.