Enriching software architecture models with statistical models for performance prediction in modern storage environments

Model-based performance prediction approaches on the software architecture-level provide a powerful tool for capacity planning due to their high abstraction level. To process the increasing amount of data produced by today's applications, modern storage systems are becoming increasingly complex having multiple tiers and intricate optimization strategies. Current software architecture-level modeling approaches, however, struggle to account for this development and are not well-suited in complex storage environments due to overly simplistic storage assumptions, which consequently leads to inaccurate performance predictions. To address this problem, in this paper we present a novel approach to combine software architecture-level performance models with statistical models that capture the complex behavior of modern storage systems. More specifically, we first propose a general methodology for enriching software architecture modeling approaches with statistical I/O performance models. Then, we present how we realize the modeling concepts as well as model solving to obtain performance results. Finally, we evaluate our approach extensively in the context of three case studies with two state-of-the-art environments based on Sun Fire and IBM System z server hardware. Using our approach, we are able to successfully predict the application performance within 20 % prediction error in almost all cases.