A Study On The Methods And Affecting Factors Of Large-Scale Load Generation

For online software systems,it is often necessary to effectively conduct load testing to ensure the reliability of their services.With the development of cloud computing technology,load testing is gradually migrated to the cloud,it is often necessary to generate large-scale load pressures in the cloud.However,the existing research on the load generation method lacks in-depth analysis and quantitative study on the factors that affect the load generation.The optimization of load generation method lacks reference,which makes it difficult to generate large-scale loads with limited test resources.For resource allocation of load generation,there exists a problem that the current methods leverage coarse-grained resource allocation with an entire virtual machine occupied by a single test task.This may easily waste the test resources and causing it is difficult to support large-scale load generation under multiple test tasks with limited test resources.To solve these problems,this thesis proposed a cloud-based framework for large-scale load testing,we quantitatively studied the main factors that may affect the load generation.This can help obtain an optimized load generation method.We also proposed a shared-mode resource allocation method to enable the sharing of cloud test host resources among load testing tasks.In more detail,the work in this thesis includes:(1)A cloud-based framework for large-scale load testing was proposed.It provides test services with different client-side cluster patterns,rich types of concurrent load mechanisms,and various forms of script languages.This can provide a variety of methods for load generation.(2)A quantitative study of the main factors that affect large-scale load generation.we analyzed these main factors.and proposed capability assessments for client load generation.We quantified the impact of load generation mechanism,hardware resource allocation and test cluster to build an optimized load generation method.This provides references for economically generating large-scale loads.(3)A shared-mode resource allocation method for cloud-based load testing was proposed.We built a multi-objective,constrained shared-mode resource allocation model.Based on this,we introduce a resource allocation algorithm to periodically allocate cloud resources for the load testing task sequence within a sliding window.The method can optimize the resource allocation of large-scale load generation.(4)We developed a system based on the cloud-based large-scale load testing framework and verified the effectiveness of the proposed methods.