Throughput-latency tradeoffs in concurrent communication and DSPsystems

摘要

The actual throughput, latency, and cost-effectiveness of feedbacksystems, concurrent recursive systems, and generalized concurrentsystems are analyzed, and an overall design methodology is proposed. Thedifferences between the throughput speedup, the iteration bound speedup,and the intrinsic throughput-latency tradeoffs in concurrent design arediscussed. It is shown that the iteration bound can be relaxed at theexpense of latency and that a concurrent system can achieve the relaxediteration bound only if the recursion hardware is not pipelined andsufficient concurrency (sometimes higher than the improvement in theiteration bound) is available