Design Of Dynamic Task Allocation Mechansim For NoC Manycore Fault-Tolerant System

The many-core system based on the on-chip network has the characteristics of low power consumption,abundant available resources,and high communication efficiency.However,there are many challenges in allocating software tasks to NoC hardware.Meanwhile,NoC Systems also face permanent failures due to manufacturing defects and aging.The NoC task allocation not only affects the performance of upper-layer software applications,but also plays an important role in improving system reliability and node fault tolerance.This paper will focus on the task allocation strategy in the many-core system,and design a high-performance dynamic task allocation mechanism and apply it to node fault tolerance.In order to improve the performance of software applications,this paper designs an online task allocation mechanism which is mainly divided into three steps: initial node selection,near convex area selection,and node mapping.The initial node selection mechanism is a proactive strategy,which can calculate the necessary variables in advance before task assignment requests,ensuring the realtime nature of initial point selection processes.Subsequently,this paper applies the ripple effect to the area selection and proposes an algorithm,that is,the allocation of the system fragmentation region is prioritized,then the utilization of the NoC system resources will be improved.Finally,a block-aware node mapping scheme is prop which maps the traffic between nodes in the software topology and the number of adjacent nodes in the hardware topology,and completes intra-area node mapping.Secondly,this task allocation is used for NoC node fault tolerance.Before failures occurred,the relationship between task allocation mechanisms and reliability was analyzed.After faults occurred,this paper proposed two faulttolerant schemes based on the number of fault cores.Both of them can complete node fault tolerance and mitigate performance degradation after fault tolerance.Finally,according to the dynamic task allocation mechanism proposed in this paper,testing and evaluation are carried out.Based on the evaluation indexes,other allocation algorithm was used as the benchmark.Experiments show that the dynamic task allocation mechanism shows better performance.Meanwhile,when the fault occurs,the limited and unlimited and task redistribution strategies proposed in this paper can be completed under the condition of node fault tolerance and mitigate performance degradation after fault tolerance.