Zero-copy Queues for Native Signal Processing Using the Virtual Memory System

The high performance of current general-purpose processors makes it feasible to perform digital signal processing on the main CPU of a workstation. As the performance gap between CPU speed and memory speed continues to increase, it becomes obvious that DSP applications on workstations must address memory performance. We present a technique for using the virtual memory system to implement zero-copy queues. We present benchmark results for a frequency-domain FIR filter using this technique, compared to manual data circularity management. The technique presented can significantly reduce the overhead of data management when processing continuous, overlapping streams of data.