Traffic dependencies in client-server systems and their effect on performance prediction

Client-server systems are becoming increasingly common in the world today as users move from centralized mainframe facilities to networks of distributed workstations. This form of work demands new performance models as the interactions in client-server systems are more complete than the types supported by classic queueing network solvers such as Mean Value Analysis. However, certain interaction patterns can arise in multi-level client-server systems that require special treatment. The paper describes these interactions (referred to as interlocking here) and how they affect the performance estimates of solution methods using surrogate delays to solve multi-level client-server models. It then describes a method to take interlocking into account when solving the performance models. These corrections often reduce the solution error to close to zero when compared to exact solutions for situations where interlocking is significant