CDI-Type I: Accelerating Simulations Using CPU+FPGA Heterogeneous Processing

生物和生理过程的计算建模正在成为任何旨在理解复杂系统的综合研究工作的一个日益增长的组成部分。这类研究的一个重要限制是用于评估通常跨越多个时间尺度的复杂数据集的计算效率。其中一个例子是心脏，它是身体中最重要的器官之一，对于它来说，计算机建模对于理解其功能变得越来越重要。除了心脏中的大量细胞之外，当考虑许多心脏过程中固有的反馈（需要精细的时间尺度）和各种刺激的长期影响（需要长时间模拟）时，心脏功能的计算建模和模拟变得极其复杂。因此，需要进行多时间尺度的模拟。该项目将利用心脏功能作为模型系统来开发新的更快的计算工具。将开发新的处理能力，用于在现在商品计算机系统中变得司空见惯的异构处理单元之间自动划分时间步内和时间步间计算。特别是，计算将被划分并映射到图形处理单元（GPU）和现场可编程门阵列（FPGA）以及通用中央处理单元（CPU），每个处理单元针对不同类型的并行性和计算瓶颈。这些能力将大大加快多时间尺度模拟和其他计算复杂的算法，提供对心脏功能和其他复杂系统的更好理解。该项目将产生一个开源资源，用于将计算划分为异构计算单元，这将在一系列学科中广泛使用。这项研究将为系统生物学和计算机科学的研究生提供跨学科的培训。除了开展该项目的大部分研究外，这些学生还将开展旨在鼓励年轻学生从事科学和工程职业的外联活动。

The computational modeling of biological and physiological processes is becoming a growing component of any comprehensive research effort aimed at understanding complex systems. A significant limitation to such studies is the computational efficiency used to assess complex data sets that often span multiple time scales. An example of this is the heart which is one of the most important organs in the body, and for which computational modeling is becoming increasingly important to understanding its function. In addition to the large number of cells in the heart, computational modeling and simulation of cardiac function becomes extremely complex when considering both the feedback inherent in many cardiac processes (requiring a fine timescale) and the long-term effects of various stimuli (requiring long simulations). Multi-timescale simulations are therefore required. This project will utilize cardiac function as a model system to develop new and faster computational tools. New processing capabilities will be developed for automatically partitioning intra- and inter-time-step computation among heterogeneous processing units that are now becoming commonplace in commodity computer systems. In particular, computation will be partitioned and mapped to graphics processing units (GPUs) and field programmable gate arrays (FPGAs) in addition to general-purpose central processing units (CPUs), with each processing unit targeting different types of parallelism and computational bottlenecks. These capabilities will dramatically accelerate multi-timescale simulation and other computationally complex algorithms, providing an improved understanding of cardiac function and other complex systems. The project will produce an open-source resource for partitioning computation to heterogeneous computing units that will be broadly useful across a range of disciplines. The research will provide interdisciplinary training for graduate students in systems biology and computer science. In addition to carrying out much of the research on this project, these students will also perform outreach aimed to encourage younger students to pursue careers in science and engineering.