摘要：Multi-tenancy application is one of the important resource sharing approaches which enables sharing of resources across a large pool of users. A key requirement to manage the resources fairly among tenants is that one should know the impacts of the resource allocations strategies. Performance model has the native advantages to predict the impacts in advance. However, as the platform become more difficult than ever before, it is hard to build the model manually. In this paper, we provide a configuration driven modeling approach to automatically build the configurations’ performance impacts. In our work, we firstly decouple the impacts into several templates which can be instanced with the runtime configurations. And then, these models can be merged with the tenant’s model as a holistic performance model, which can be used to predict performance under different configurations. Experimental results derived from testing the approach by using an online application deployed on a widely-used platform illustrate the potential of this approach.
