Load balancing and stability issues in algorithms for service composition

Service composition enables flexible creation of new services by assembling independent service components. We are focused on the scenario where such composition takes place across the wide-area Internet. We envision independent providers deploying and managing service instances and portal providers composing them to quickly enable new applications in next-generation networks. One of the important goals in such service composition is load balancing across service instances. While load balancing has been studied extensively for web-server selection, the presence of composition presents new challenges. First, each client session involving composition requires a set of service instances and not just one server. Second, unlike web-mirror selection, we also concern ourselves with load balancing in the presence of failure recovery during a client session. We introduce (a) a metric to choose the set of service instances for composed client sessions: the least-inverse-available-capacity (LIAC) metric, as well as (b) a piggybacking mechanism to give quick feedback about server load. We then introduce an additional factor in the load balancing metric to avoid choosing far away service instances. Our experiments, based on an emulation testbed, show that our load balancing mechanism works well under a variety of scenarios including network path failures.