The Research Of Fault Tolerance Technology For Real-time Systems Based On The Hierarchical Scheduling

In recent years, real-time systems have been widely used in the safety-criticalfield of automotive electronics. In addition to ensure the logical correctness of outputresults, real-time applications also must have a strict time certainty and a high degreeof reliability. Otherwise it will lead to disastrous consequences. However, with theexpansion of application requirements, the current real-time application softwares aredevelopping in the direction of large-scale and high complexity. Its security andreliability problems become increasingly prominent. It is very necessary to make useof fault-tolerant technology to control the behavior of real-time systems after an error.This paper introduces a knowledge of the hierarchical scheduling framework.Then it gives a detailed analysis of the fault-tolerant methods based on taskreplication which have widely used in distributed real-time system. The problem ofthese methods could be summarized thus:1) the existing system reliability modelassumes that the system only exists a failure at a certain time, and it has been solvedwhen the next failure is coming. It is limited in practical application;2) these methodsalways blindly make each task with ε+1replicas to tolerance ε possible faults, whichis in order to improve the reliability of the system. However, it comes at the expenseof causing the high redundancy of the system, and could lead tasks to miss theirdeadlines owing to the competition of computing resources.With the help of the hierarchical scheduling framework and the fault-toleranttechnology based on the active task replication, we design improveed strategies forthese shortcomings, aimed at meeting the reliability targets and minimize redundantresources. Then we expand the research of fault-tolerant theory and algorithms forcomplex embedded real-time systems on multiprocessor platform.First of all, a task replication algorithm for periodic tasks in a homogeneoussystem is proposed. In the modeling phase, we use the method of probability andstatistics to analyze the reliability of the whole system in a super-period. Sequentially,the reliability model for the real-time system based on periodic task set isimplemented. Secondly, it is the precisely quantitative phase. On the basis ofreliability targets and the analysis of the system’s reliability model, we give themethod to calculate the lower limit value of replicas. It can avoid blindly taskreplication to some extent. Then, by exploiting the difference of task resourceoverhead and its contribution to the reliability of the system, we put forward a kind of economic task replication strategy. It can dynamically establish the number ofreplicas for each task. Meanwhile, it would reduce the increasing consumption ofcomputing resources due to task replication as far as possible.Secondly, we put forward a kind of task replication algorithm for DAG tasks in aheterogeneous system. In view of the modeling phase, the effective reliability modelof single DAG figure is firstly established through analyzing the dependenciesbetween the tasks. And base on it, we implement the modeling of reliability for themuti-DAG system. In the quantitative analysis phase, a algorithm of determining thelower limit value of replicas is suggested with the aid of the established reliabilitymodel of the whole system and the thought of selecting the processor owned theminimum reliability cost each time. Then under the driven of the system’s reliabilitytarget, we adopt the economic task replication strategy above to quantify the numberof replicas needed for every task dynamically. And all the replicas are allocated to thecorresponding processors on the multiprocessor platform.The simulation results show that, compared with the previous blind fault-tolerantmethod, our proposed algorithms could effectively guarantee reliability targets andminimize resource consumption simultaneously.