A Language Independent Framework for Compiling Data-Intensive Applications on Highly Parallel Systems

最近，术语数据密集型超级计算(DISC)越来越流行，包括在海量数据集上执行大规模计算的应用程序。由于分析的数据量越来越大，涉及的计算量越来越大，以及对快速甚至交互响应的需求，这些应用程序对计算能力的需求越来越大。从最近2-3年开始，不再可能通过简单地增加时钟频率来提高处理器性能。因此，现场可编程门阵列(FGA)和图形处理单元(GPU)等多核架构和加速器已成为提高性能的经济有效的手段。然而，这样的体系结构给这类应用程序带来了可编程性挑战。该项目的目标是一个独立于语言的编译器和运行时框架，使数据密集型应用程序能够在各种现代和新兴的高度并行系统上进行扩展。具体地说，目标将是多核计算机集群，其中每个节点都可以额外拥有一个类似于GPU的加速器。这里提出的系统将建立在我们之前在更早的系统Freeride(数据挖掘引擎的快速实现框架)上的工作基础上。在Freeride框架的基础上，该项目有可能在高端计算、数据挖掘和科学数据处理领域产生影响。

Recently, the term Data-Intensive SuperComputing (DISC) has been gaining popularity and includes applications that perform large-scale computations over massive datasets. Because of the increasing volume of data analyzed, the amount of computation involved, and the need for rapid or even interactive response, these applications have an ever increasing demand for computational power. Starting within the last 2-3 years, it is no longer possible to improve processor performance by simply increasing clock frequencies. As a result, multi-core architectures and accelerators like Field Programmable Gate Arrays (FPGAs) and Graphics Processing Units (GPUs) have become cost-effective means for scaling performance. Such architectures are, however, creating a programmability challenge for this class of applications. This project targets a language-independent compiler and runtime framework for enabling data-intensive applications to be scaled on a variety of modern and emerging highly parallel systems. Specifically, the target will be cluster of multi-core machines, where each node could additionally have an accelerator like a GPU. The system proposed here will built on our prior work on an earlier system, FREERIDE (FRamework for Rapid Implementationof Datamining Engines). Building on the FREERIDE framework, this project has the potential for an impact in the areas of high-end computing, data mining, and scientific data processing.