Optimization Research On Job Scheduling Of Mass Seismic Data Parallel Processing

MPS is suitable for marine seismic data parallel processing system of PC cluster, is themassive seismic data processing and development platform to solve the problem ofcomplicated calculation. It established a parallel job scheduling system, has the basic functionof the job scheduling management. But the system has many deficiencies, such as: schedulingalgorithm of job scheduling module also has the flaw, seismic job processing efficiency is notsatisfactory, etc.In order to solve the above problems, this thesis through studying the architecture of theseismic data job scheduling system, analyses the processing procedure of seismic job,optimizes job selection algorithm and node allocation algorithm of job scheduling module,and puts forward the model of job parallel processing. In order to improve the job executionand use of system resources more effectively, it optimizes the I/O scheduling of jobprocessing. This main research content is as follows:1. Research job scheduling system architecture and related technologiesResearch the composition of job scheduling system, analyses the job scheduling moduleand its deficiency in job process, put forward the improved scheme of job scheduling moduleand job processing model. In addition, master-slave mode and master-slave node function ofthe seismic data job is introduced in the thesis.2. Research job scheduling algorithm and node allocation algorithmHaving analyzed the principle of current job scheduling algorithm, aiming at thedeficiency of the algorithm, improved backfill scheduling algorithm was proposed. Thealgorithm through modifying the kernel number of small job of job waiting queue, makes itequal to the kernel number of cluster idle resources, makes little job that has no chance ofbackfill get chance of backfill. At the same time, the nodes allocation algorithm is optimized,and through reducing the proportion of job outside communication the job execution isimproved.3.Research job processing model and I/O scheduling optimization of seismic dataprocessing system (MPS)This thesis through analyzing job process of seismic data processing system (MPS), putsforward the seismic data job parallel processing model. The model implements dividing inputmodule, processing module, output module into three parallel process, gives full play to theadvantage of multi-core processor, and then improves job execution. In addition，through researching the current seismic data extract trace set methods, it analyzes the defects of thecurrent seismic data extract trace set method, optimizes the current seismic data extract traceset methods. when the seismic data job reads the file, it’s as far as possible to lower thenumber of reading disk and positioning disk, so as to improve the efficiency of the seismicdata reading.Finally, testing the optimized system, it calculates the average response time, the averagerunning time and resource utilization of the test jobs. through comparing with beforeoptimization shows that the optimized scheduling algorithm improves the cluster resourceutilization, it reduces the running time of the job.