Accelerating the continuous data in a SCI-based multimedia system

The SCI (Scalable Coherent Interface) is widely used in high performance clustered systems these days. However, there has not been any consideration for scheduling policies of improving transmission performance of continuous data in a SCI-based multimedia server. The scheduling policy that does not take into account the property of continuous data may hurt the overall performance. We propose two scheduling policies. The first policy is to maintain separate queues in a node in order to provide differentiated transmission services to continuous data and discrete data. The second one is to let continuous data even preempt discrete data that is being transmitted, which may bring more timely service to continuous data. We measured how the continuous data latency is reduced by simulations, which show that 8-70% reduction is expected, on an average, with scheduling policy and observed even better reduction with the second scheduling policy.