Multi-tenant Application Resource Consumption Research Based On Regression Analysis

Multi-tenant architecture allows tenant users sharing the same infrastructure so that it can reduce cost. It is one of the key features of cloud computing. Because of the differences of resources demand and changing workload, the tenant’s consumption of resources is also not identical. In multi-tenant environment, since the problem of security issues and the isolation issues, we need to obtain resource consumption real-time. Due to the application or system environment, it is not allow to modify the environment, so direct measure is not suitable. Based on the analysis method of multi-tenant applications, we use a mathematical method to discuss the multi-tenant application resource consumption.Firstly, direct measurement used to measure tenant’s application resource consumption of memory, network and so on. However, due to the tenants’ sharing environment, CPU consumption is not easy to obtain. Due to the problem of overhead, direct measurement method is not suitable for multi-tenant environment. The traditional analysis methods of mathematics in a multi-tenant environment because of effects of collinearity caused by changing workload, the estimation result is not good. The traditional method is not suitable for multi-tenant for the sharing of resources and dynamic workload environment, so we proposed a method for CPU consumption estimation based on the regression analysis method. Considering workload, response time, average consumption time and transaction throughput to predict CPU consumption. Experiments show that, compared with the traditional regression analysis method to estimate the CPU consumption, this method can effectively improve the accuracy.Secondly, based on obtain resources consumption of tenant, we analysis the workload and complete the prototype multi-tenant resource adjustment system. There is problem of excessive resources competition. The high workload could cause problem of longer response time, longer waiting time and lower quality. So we predict the workload by analysis the history workload and dynamic allocation resource in the prototype system.In this paper, we focused on CPU consumption and proposed a new regression analy sis strategy to improve the accuracy. Then, based on virtualization, we analysis the histori cal workload information and predict, then use it for multi-tenant resource adjustment prototype system.Test results show that, in continuously changing workload environment, th e proposed strategy adapts to the multi-tenant environment well, the multi-tenant resource a djustment prototype system could prevent overhead, so that improve the resource utiliz ation rate.