Theoretical construction and application of fine-grained cycle sharing technique for GPU grid

我们为计算网格系统开发了一种细粒度的循环共享技术，共享网络上的计算资源。我们的技术基于图形处理单元(GPU)的多任务机制。我们将我们的技术应用于生物信息学问题，称为同源搜索，并证明了在编辑电子文档的同时可以加速科学应用。我们还构建了GPU的空闲周期模型，并提出了一种根据空闲周期长度的分布来选择计算资源的方法。

We developed a fine-grained cycle sharing technique for computational grid systems, which share computational resources on the network. Our technique is based on a multitasking mechanism for graphics processing units (GPUs). We applied our technique to a bioinformatics problem, called homology search, and demonstrated that scientific applications can be accelerated while editing electronic documents. We also constructed an idle cycle model for GPUs and developed a method that selects computational resources according to the distribution of idle period lengths.