Thumbnail Image

Traffic Prediction for NoCs using Fuzzy Logic

Thomas, Gervin; Juurlink, Ben; Tutsch, Dietmar

Networks on Chip provide faster communication and higher throughput for chip multiprocessor systems than conventional bus systems. Having multiple processing elements on one chip, however, leads to a large number of message transfers in the NoC. The consequence is that more blocking occurs and time and power is wasted with waiting until the congestion is dissolved. With knowledge of future communication patterns, blocking could be avoided. Therefore, in this paper a model is introduced to predict future communication patterns to avoid network congestion. Our model uses a fuzzy based algorithm to predict end-to-end communication. The presented model accurately predictions for up to 10 time intervals for continuous patterns. Communication patterns with non-continuous behaviors, such as fast changes from peak to zero, can also be predicted accurately for the next 1 to 2 time intervals to come. The model is a first step to predict future communication patterns. In addition, some limitations are identified that must be solved in order to improve the model.
Published in: High-performance and hardware-aware computing : proceedings of the Second International Workshop on New Frontiers in High-performance and Hardware-aware Computing (HipHaC'11), San Antonio, Texas, USA, February 2011 (in conjunction with HPCA-17), KIT Scientific Publishing
  • Proceedings DOI: 10.5445/KSP/1000021732 (https://doi.org/10.5445/KSP/1000021732)