Research On The Earthquake Simulation Oriented Scheduling System

Now China is in the period of rapidly developing economic and society, it is of great requirements to reduce the harm of earthquake disaster. It plays a great role to establish an integrated earthquake disaster simulation system for reducing the impact of the earthquake. In the field of earthquake disaster simulation and prediction, there are very huge computing needs, but the volume of data needed in computation is quite large, and the computing cycles is so long that the high-performance supercomputers are required. However, as the computer and network technology continues to evolve, due to the high price and low scalability of supercomputers and also due to the inadequate of one computer on large-scale co-simulation, the limitations of the traditional method is becoming more evident.Therefore, this paper combines the distributed computing theory with the needs of earthquake disaster simulation, targets to study the scheduling system orienting earthquake simulation, analyses the task scheduling heuristic, the method of related software design, and the simulation for heuristics. Generally speaking, it mainly includes:Based on the heuristics HLTF(The Heterogeneous Largest Task First) and Segmented Min-Min, we propose HLSF heuristic to deal with mapping of several independent tasks first by grouping them in to meta-task, and then assigning them to machines. The HLSF heuristic first introduces an n-1 dimensional vector to describe the machine heterogeneity, and then the machine heterogeneity of each sub-task is reflected in hasse diagram. After this, according to the level of the tasks in the hasse diagram, they are mapped with the top-down approach. In this way, the heuristic combines the heterogeneous largest tasks first and the method of segments, and also it corrects the arbitrary partitioning of the matrix in Segmented Min-Min and overcome the lack of an effective measure of heterogeneity in the HLTF algorithm. Experimental results illuminate that our heuristic outperforms the traditional ones.Actually, the execution time of some tasks is very long, so we develop a simulation environment based on simjava, compared with traditional simulation tools, such as simgrid, gridsim, which can only generate the consistent ETC (expected time to compute) matrices, our simulation environment generates the ETC matrices in the cov-based (coefficient-of-variation-based) method which provides greater control over spread of the execution time values than the range-based method used widely previously.Finally, we develop a scheduling system CqScheduler prototype which orients earthquake simulation, by using HLSF heuristic, the makespan of simulation tasks can be ensured smaller.