Buffer Sizing for Multimedia Flows in Packet-Switching NoCs

Abstract

Wormhole packet switching, used in many NoC designs, introduces jitter. This may produce violations of application deadlines. Several works in the literature propose stream workload models, and techniques for buffer sizing. These works do not consider the concurrency between different flows, or the NoC model is too abstract, masking the jitter introduced by data packaging and router processing. One technique to deal with jitter is to introduce a decoupling buffer (D-buffer) on the target IP. This buffer receives data from the NoC with jitter, while the target IP consumes data from this buffer at the application rate, without jitter. Two problems must be solved to implement D-buffers: (i) which size must the buffer have? (ii) how much time should be expected before data consumption starts (threshold)? This work proposes a general method to define D-buffer size and threshold, considering the influence of packaging, arbitration, routing and concurrency between flows. A traffic model for stream applications is detailed and used to characterize jitter sources in wormhole packet switching. Experimental results demonstrate the impact on multimedia flows with fixed and variable packet sizes (from real traffic traces). Simple traffic models employing constant frame sizes result in small D-buffers. On the other hand, employing multimedia frames from application traces (i.e. real application data) increases buffer size and threshold while still suppressing jitter. Another parameter analyzed in the paper is the percentage of deadline violations as a function of D-buffer size. The method guarantees throughput with no deadline violation and does not modify the NoC structure. The cost of adding such D-buffers is an increased latency and extra silicon area.