Research On Performance Management Mechanism Based On SLA In SaaS Application

The development of Internet technology and the maturity of application software give birth to a new model of service delivery and operation -Software as a service (SaaS). The typical character of SaaS applications is'single-instance multi-tenancy'. Specifically, scale effect suggests that multiple tenants sharing a single software instance reduce the service cost. In addition, the sharing brings about great flexibility and scalability for SaaS applications. Tenants have different requirements of services in terms of service function, service period, mean response time, supported throughput, or availability. A service level agreement (SLA) is required for the coordination between service providers and tenants. Service providers configure their resources on the basis of the SLA to meet tenants'customized requirements. In the service period, the providers also monitor the SaaS applications and optimize the service performance to meet the quality parameters of the SLA.Several problems, however, still exist in the application of the traditional SLA management. First, although SLA is able to describe heterogeneous requirements for multi-tenant oriented application with great clarity, its static characteristic indicates some tenant operating against the signed SLA. Second, the failure of any SLA control may lead to system instability and the decline of server's overall performance, which will in turn affect the service quality of other tenants. This is due to the share mechanism between different tenants. Last, the existing performance management mechanism treats every tenant equally and fails to optimize the performance according to their different SLA levels, which will further decrease the tenants' satisfaction. In a nutshell, the traditional SLA management mechanism is no longer satisfied because of the characteristics of "single-instance multi-tenancy." In this case, there's an urgent need for us to build a new SLA-based performance management mechanism.In this paper, we mainly contribute in the following aspects:1. Propose a performance management mechanism based on the SLA in SaaS application. We demonstrate that the feasibility and reliability of this implementation through a detailed explanation of the process and a design of SLA management mechanism which is adequate for SaaS application.2. Give a performance monitoring method and a violation detection proposal for SaaS. We define the violation states and aggressive tenants, and then monitor the application performance and server resource periodically. The proposal is able to detect the violation states using violation detection algorithm and infer the candidates of aggressive tenants according to the performance data obtained by the monitor.3. Propose a variety of SLA-based dynamic scheduling algorithms according to the hierarchy of the SLA application deployment, including tenant placement method and request scheduling algorithm. These algorithms improve the service performance, ensure the operation of the SaaS applications, and provide stable resources for tenants.This paper proposes a new SLA performance management model for SaaS applications through improving the SLA management mechanism for the traditional web service. This model is capable of real-time performance monitoring and multi-level performance optimization on the basis of the quantification of SLA parameters. The service quality is fully guaranteed by monitoring service grades, detecting violation status, and dynamically scheduling shared resources. In addition, this paper also verifies the feasibility and reliability of the model in terms of the functional implementation, multi-tenant support, and performance adjustment.