Study Of Real-time Scheduling For Cyclic Transaction Set In NoC

It is close to the limits that getting high computational performance by increasing the processor frequency, which cannot meet people’s requirements for high computing performance, so the multi-core becomes a new development direction. But with the increase of the number of cores, the traditional bus organized structure greatly affects the performance of the system, so the Network-on-chip (NoC) emerged. NoC make tasks in system communicate with each other in parallel, greatly enhancing the communication efficiency of the system.Due to their multiple degrees of parallelism both in computation and communication, using SoC in real-time systems requires new theoretical frameworks. Scheduling of real-time transactions in a NoC while maximizing its utilization poses new challenges, which are different from those of traditional real-time systems. In particular, although the problem has the form of multiple resource scheduling, it is not the same as the traditional multi-processor scheduling because a transaction may use multiple resources, i.e. physical links simultaneously.Previous studies on real-time communication in NoC are mostly concentrated in the acyclic communication task set schedulability analysis, while there is little research on the cyclic communication tasks set. However, the communication task set the practical application is not always acyclic. Cyclic Transaction set in NoC refers to those whose routing path overlap with each other and form a loop.In this thesis, two scheduling algorithms for the cyclic task sets are proposed: Real-time Scheduling Algorithm based on virtual channel (CS-VC) and Real-time Scheduling Algorithm based on time isolation. And we provide the utilization boundaries of transaction sets for these two algorithms (also the sufficient condition for the transaction sets to be scheduled successfully). In the experiment, we compare these two algorithms with currently the best algorithm for cyclic transaction sets scheduling. Experimental results show that CS-VC and CS-TI both have a better reception rate, and the smaller of transaction sets and the scale of NoC, the better the reception rate of CS-CV and CS-TI is. Even though CS-TI has a worse reception rate than CS-CV, it requires no more additional NoC components and is easier to be implemented.