High speed eigenvalue solver on the Cell cluster system for controlling nuclear fusion plasma
High speed eigenvalue solver on the Cell cluster system for controlling nuclear fusion plasma
复制标题
用于控制核聚变等离子体的细胞簇系统上的高速特征值求解器
DOI：
10.1109/pdp.2010.22
复制
发表时间：
2010
期刊：
影响因子：
--
通讯作者：
and Shinji Tokuda
中科院分区：
文献类型：
--
作者：
Noriyuki Kushida;Hiroshi Takemiya;and Shinji Tokuda
在本研究中，我们开发了一种高速特征值求解器，它是国际热核实验反应堆（ITER）在细胞簇系统上等离子体稳定性分析系统所必需的。我们的稳定性分析系统是为了防止等离子体破坏对ITER的破坏而开发的。MARG2D是该系统的主要组成部分，能够实时分析等离子体的状态和被测条件。根据我们的估计，MARG2D算法中最耗时的部分是特征求解器，我们必须在1秒内求解出100000维的结果特征系统。然而，目前的大规模并行处理器（MPP）类型的超级计算机并不适用于这种瞬时计算，因为网络通信的开销占主导地位。因此，我们采用Cell集群系统，它的处理器性能比MPP更高，因为我们可以用较少的处理器获得足够的处理能力。此外，我们还考虑了Cell集群的层次结构：处理器间、处理器内和SIMD并行性，开发了新的特征值求解器。最后，我们在1秒内成功解出了块的三对角线厄米矩阵，该矩阵有1024个对角线块，每个块的大小为128 x 128。
In this study, we developed a high speed eigenvalue solver that is the necessity of plasma stability analysis system for International Thermo-nuclear Experimental Reactor (ITER) on Cell cluster system. Our stability analysis system is developed in order to prevent damages to the ITER from plasma disruption. MARG2D, which is the main part of the system, analyzes the state of plasma with the measured conditions instantaneously. According to our estimation, the most time consuming part of MARG2D is eigensolver and we must solve resulted eigensystem whose dimension is hundred thousand within a second. However, current massively parallel processor (MPP) type supercomputer is not applicable for such instantaneous calculation, because the overhead of network communication becomes dominant. Therefore, we employ Cell cluster system, whose processor has higher performance than MPP’s, because we can obtain sufficient processing power with small number of processors. Furthermore, we developed novel eigenvalue solver with the consideration of hierarchical architecture of Cell cluster: Inter processor, intra processor, and SIMD parallelism. Finally, we succeeded to solve the block tridiagonal Hermitian matrix, which had 1024 diagonal blocks and the size of each block was 128 x 128 within a second.