Optimal Replication for a Specific System Objective

This paper defines component replication as the concept of using two or more i.i.d. components at a specific location in a system design. Both parallel and series physical connections are used to place replicate components into a system. The choice of parallel or series depends upon why the components are being made replicate and is complicated by the fact that a component might be used for several functions. A given physical connection can be logically in series for some situations and logically in parallel for others. Reducing system unavailability is one important use of component replication. The mathematical models relating replication and unavailability are often complex, and casual inspection does not reveal which components should be replicated to reduce unavailability. This paper develops a method for determining which components can be replicated (and how) for the greatest unavailability reduction. The method was used to study three system designs. For each design it was possible to identify several components which could be replicated with an associated reduction in unavailability at reasonable cost. In addition it was possible to demonstrate or confirm that replicating most components does not reduce unavailability. Applying this method helps the designer develop insight concerning the role of replication (in a system design) without complex calculations and interpretations.