Service address routing: a network-embedded resource management layer for cluster computing

Service address routing is introduced as a novel and powerful paradigm for the integration of resource management functions into the interconnection fabric of cluster computers. SAR provides a “location independent” mechanism for the distribution of computations (services) among the computational resources of the cluster. The intelligence to allocate services and, later on, invoke their instantiation is embedded into intelligent switching devices [Isaac D. Scherson, C.-K. Chien, Least common ancestor networks, VLSI Design 2(4) (1995) 353–364]. Invocation of services is effected transparently to requesting nodes by these network-embedded management functions. Thus, applications can benefit from the inherent parallelism of the cluster while being totally unaware of the specific “location” where required services are rendered. The performance of SAR’s service discovery mechanism in hierarchical Least Common Ancestor Networks is studied by means of simulating two different system configurations: level-global knowledge and Level Caches. It is shown that searches using Level Caches work better than level-global knowledge for a typical scientific computing workload.