Two Optimization Techniques for Component-Based Systems Deployment

Performance of a distributed system can be greatly affected by the deployment of the system components over the nodes of the target network. Performance evaluation and tuning after deployment is generally expensive and risky because redeploying the system is a complex task. We present two approaches to optimizing deployment of component-based systems: a binary integer programming model (BIP) and a multiobjec-tive niching genetic algorithm (MNGA). We demonstrate the two approaches by using them to deploy instances of an implementation of a distributed replicated data service called Eventually Serializable Data Service (ESDS). The interesting aspect of the work is that while BIP model guarantees the optimal deployment according to the chosen criterion, a good GA solution can be obtained in a much shorter time.