Throughput unfairness in fair arbitration interconnection-buses for aerospace embedded systems

Abstract

Currently, embedded systems are composed of processors, memories, and Intellectual Property Cores (IP Cores) interconnected to develop a set of specific tasks. Therefore, the selection of an appropriate interconnection architecture is critical in terms of system performance and functionality. A Network-on-Chip provides an efficient and scalable interconnection solution when there are a large number of elements in the system. However, the bus-based interconnection system remains the best option to connect a few cores. The bus arbiter uses an allocation policy to select which IP Core obtains access to the bus. The so-called fair policies ensure that all processors in the system have the same opportunity to access the bus. However, they fail to offer a fair share of the bandwidth or transmission rate, especially when there are heterogeneous IP Cores. As a study case, we analyze an embedded aerospace system for earth observation. Different IP Cores preprocess satellite images at distinct execution times -and unbalanced processing ratesaffecting the delivery rate of images to earth. We study the phenomenon of uneven bus transmission rates due to improper bus allocation using policies such as Round Robin, FIFO, and Lottery. Also, we propose a metric to compute the maximum number of IP Cores without bus saturation.