Simulation Metamodeling for the Design of Reliable Object Based Systems

Replication is a suitable approach for the provision of fault tolerance and load balancing in distributed systems. Object replication takes place on the basis of well-designed interaction protocols that preserve object state consistency in an application transparent manner. The published analytic performance models may only be applied in single-server process replication schemes and are not suitable for schemes composed of miscellaneous policies, such as those arising in object based systems. In this work we make use of a simulation metamodeling approach that allows the comparative evaluation of composite fault tolerance schemes, on the basis of small size uniform experimental designs. Our approach opens the possibility to take into account different design concerns in a combined manner (e.g. fault tolerance combined with load balancing and multithreading). We provide results in terms of a case system study that reveals a dependence of the optimal adjustments on the system load level. This finding suggests the device of dynamically adjusted fault tolerance schemes.