Research On Key Techniques Of Radio Resource Allocation In Relay-Enhanced Cellular Networks

The Hadoop distributed system framework solves the problem of data storage and data processing effectively at the age of big data, it’s performance is closely to the job scheduling. Adopting more efficient scheduling scheme, we can allocate and use of the cloud computing resources more rationally. Obviously, the study about the job scheduling algorithm of Hadoop technology has important practical significance to improve the overall performance of the Hadoop system.This paper analysis of Hadoop existing job scheduling algorithm detaily, and samely operate to the LATE scheduling algorithm which was proposed for heterogeneous environment. The focus of the paper is optimizing the defects about the speculative execution of the LATE scheduling algorithm.To the large Hadoop cluster, the probability of data storage across frame is higher. LATE scheduling algorithms assume that most of the Map task execution by read data on the local, and hasn’t consider whether the reduce tasks execution by read data on the local, this cause the time consumption in data transmission between the rack. In view of the defects that the algorithm without considering the data locality in the speculative execution, the fourth chapter puts forward an improved scheduling algorithm based on the LATE algorithm which consider the data locality. Based on the LATE algorithm, the improved algorithm improved the judge methods to the slow task in the speculative execution, and when assign the backup tasks to the slot, it will first consider the self-machine rack, if not find,then find the property backup task on other racks, this method improve the execute efficiency of the system.To the actual large Hadoop cluster environment, it is a common case to have I/O-bound and CPU-bound jobs, which demand different resources. The existing Hadoop scheduling algorithm is not aimed at the improvement of the two kinds of jobs, this cause the system resource competition still serious. Focus on the defect, the fifth chapter puts forward an improved scheduling algorithm based on the LATE algorithm, the algorithm fully consider the resources complementary allocation. Based on the LATE algorithm, the algorithm reasonable classify the load of the job and the load of the cluster’s node. The algorithm also fully consider that the CPU and I/O system resources needed are complementary, then scheduling and executing reasonably to the backup tasks. This method greatly improved the low system performance problem caused by the resources competition and unreasonable scheduling.In order to verify the advantage of the algorithms, we construct Hadoop cluster for simulation, the fourth chapter simulate two racks, and the fifth chapter use simple cluster to keep variables control.The experiment results show that this algorithm has certain advantages to the Hadoop cluster.