面向Yarn规范的蛋白质折叠模拟计算并行化算法

为了提高生物信息学中蛋白质折叠模拟计算的速度,提出了面向Yarn(Yet Another Resource Negotiator)规范的蛋白质折叠模拟计算并行化算法Yarn_PERM。分析了蛋白质折叠的格点模型PERM算法的运行流程及其面向Map-Reduce的子任务划分方式。Yarn_PERM算法实现采用Hadoop2.0的Yarn框架作为工作平台,其资源的分配与调度、应用子任务的申请和子任务的具体执行都由Yarn来透明的完成；描述了Yarn_PERM算法的Map程序与Reduce程序及主控程序的功能实现。选择了一个有代表性的蛋白质序列数据作为案例程序进行了测试。实验结果表明：在相同的时间内Yarn_PERM比PERM串行计算、Map-Reduce的PERMS计算在能量最低寻优的吞吐量上明显增加,加速比和可扩展性上也有明显的优势。

A Yarn Oriented Protein Folding Simulation Computing Parallel algorithm

In order to improve the computation speed of protein folding in bioinformatics, a Yarn oriented protein folding simulation parallel algorithm named Yarn_PERM was proposed in this paper. The computation flow of PERM algorithm and the subtask partition for Map-Reduce approach in Yarn_PERM were discussed subsequently. The Yarn framework of hadoop2.0 was used for the protein folding working Platform in Yarn_PERM. All the Resource distribution and scheduling, subtask applying and subtask executing was completed transparently in Yarn. The Map program and Reduce program for Yarn_PERM were also described in detail. In order to demonstrate the efficiency of Yarn_PERM, the simulation experiments on a representative protein data were done. The results and performance analysis demonstrated that the finding lowest energy function had been completed in a more throughput numbers than PERM and Map-reduce PERM. Good speed and scalability had been also obtained in our test.