Checkpoint Optimization Based On Active Varialbe Analysis In OpenMP Programs

Failure is inevitable in software.Software fault tolerance is increasingly important.When parallel execution sections in a OpenMP program exist faults,we can use the method of parallel recomputing to recovery faults.To realize parallel calculations,checkpoint in the program during normal operation is needed,saving the variable's calculation state.Saving all the variables will cause the problem of a large amount of storage and overhead.Omitting variables,which are not needed to store,is an effective way to reduce the saving quantity in the checkpoint.Checkpoint optimization technologies based on active variable analysis are studied in this paper,the main work is as follows:Values spreaded between implicit value and implicit reference may change,which may cause the error value to be propagated,and thus it may cause the problem that active variable values are not correct.In this paper,a series of methods,such as making variables,step-by-step implementation,adding the assignment operation and so on,are proposed int this paper and can deal with the problem,excluding the adverse effect of the implicit value and implicit reference on the active variables analysis,and the effectiveness of these methods is verified by experiments.Do research on the method of saving array data by expression in the checkpoint.Existing checkpointing method does not consider the semantic,and the copying of the original value to save the array element active variable values causes the problem of large amount of data storage.According to the characteristics that the array elements are stored orderly according to the array subscript and the first address of array elements,the expression saving method in this paper uses the expression to store the values of array element active variables,in order to reduce the data volume.And the effectiveness of the method is verified by simulation analyses.OpenMP program checkpoint optimization algorithm is proposed in this paper.Active variable analysis method,the implicit value and implicit reference treatment method and the expression method to express array data are used to reduce the amount of data in the OpenMP program checkpoint.Before live variable analysis,the algorithm identifies the implicit value and implicit reference,and then removes the adverse effect of implicit value and implicit reference on active variables analysis.Then the method of active variables analysis is used,to obtain the active variables which are needed to be saved in the checkpoint.For array element active varables,use the expression,which costs less amount,to save the value of array element active varables,in order to reduce the data storage amount of array element values.By using algorithm in NPB OMP 3.2 test procedures,through data analysis and simulation analysis,it is proved that the algorithm is effective in reducing the storage overhead of array elements.