Locations for Performance Ensuring Admission Control in Load Balanced Multi-tenant Systems

In the context of Software as a Service offerings, multi-tenant applications (MTAs) allow to increase the efficiency by sharing one application instance among several customers. Due to the tight coupling and sharing of resources on all layers up to the application layer, customers may influence each other with regards to the performance they observe. Existing research on performance isolation of MTAs focuses on methods and concrete algorithms. In this paper, we present concerns that are raised when serving a high amount of users in a load balancing cluster with multiple MTA instances. We identified potential positions in such an architecture where performance isolation can be enforced based on request admission control. Considering various approaches for request-to-instance allocation, our discussion shows that different positions come along with specific pros and cons that have influence on the ability to performance-isolate tenants.