An Extensible Message-Oriented Offload Model for High-Performance Applications

In this paper, we present and verify a new model designed to capture the benefits of protocol offload in the context of high performance computing systems. Other models capture the benefits of offload or the performance of parallel applications. However, the extensible message-oriented offload model (EMO) is the first model to emphasize the performance of the network protocol itself and models it in a message-oriented rather than flow-oriented manner. EMO allows us to consider benefits associated with the reduction in message latency along with benefits associated with reduction in overhead and improvements to throughput. In order to verify EMO, we use the tool to model a very common offload technique, interrupt coalescing. We discuss the assumptions of our model and show the modeled offload and latency performance of interrupt coalescing and no interrupt coalescing. We then present preliminary results to verify that our model is accurate. Los Alamos Computer Science Institute SC R71700H29200001