权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

A Study of Parallel Distributed Processing Kernel Ware by a Principle of the Fine-grain Multi-thread Processing

基于细粒度多线程处理原理的并行分布式处理内核研究

基本信息

批准号：
15200002
负责人：
AMAMIYA Makoto
金额：
$ 33.53万
依托单位：
KYUSHU UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (A)
财政年份：
2003
资助国家：
日本
起止时间：
2003 至 2005
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-15200002/
关键词：
Fine-grain Multi-thread Parallel Distributed Processing Kernel Ware FUCE Architecture Thread Control Mechanisms FPGA Operation System Multi Agent 細粒度マルチスレディングオーネルウェア

项目摘要

We developed the fine-grain multi-threading processor ; the Fuce architecture for parallel-and-distributed processing. The Fuce processor executes multiple threads exclusively. The core concept of Fuce is continuation-based multithread execution.In this project, we conducted the following research :・The continuation-based computing model was developed as a basis for thread level parallel computation. C-like programming language, Fuce-C, and its compiler system was developed to support to program the continuation-based thread level parallel processing.・The Fuce processor was designed and implemented on FPGA. The performance of the Fuce processor was evaluated using Fuce emulator on FPGA and software simulator.・The operating system, Cefos, was designed on the basis of continuation-based fine-grain multithreading, and evaluated its possibility using several multi-processing benchmarks.Through these researches, we showed the effectiveness of Fuce architecture. We implemented the FUCE proce … More ssor as an emulator machine on FPGA using VHDL and software simulator using Java. We evaluated several benchmarks, e.g, Prime-number by sieve algorithm, FFT, Quick sort, Merge sort and N-queen problem, with the FPGA emulator and software simulator. By this evaluation, we showed that the fine-grain multithreading is very effective for various kinds of parallel computations ; data parallel processing, pipeline parallel processing and their mixture. Especially, continuation-based multithreading technique is quite useful for high performance stream processing that is an important processing scheme for constructing OS kernel. As a result of the evaluation, we conclude the continuation-based fine-grain multithreading is a very effective method for solving the parallel processing problems such as OS kernel construction which requires basic concurrent and parallel processing techniques. We also evaluated the hardware cost of the FUCE processor and showed that the Fuce processor is hardware-cost effective compared to the commodity processors. We developed the OS kernel mechanisms based on the continuation-based fine-grain multithreading programs. And we showed the OS kernel could be constructed simple using the continuation-based fine-grain multithreading technique. We conducted a research on new architecture and basic mechanism of co-operation between hardware and software ; the kernel-ware, for parallel and distributed processing, and we showed that the Fuce kernel-ware is feasible developing the practical parallel processor. Our purpose in this research was to study the basic technology for developing a practical parallel-and-distributed machine, and we can now conclude that this research purpose was achieved. We will work further to make the Fuce machine more practical. We develop the Fuce-C language system and Cefos operating system on FPGS emulator and software simulator that runs on the commodity cluster machine. Then, we will evaluate the Fuce architecture using more scalable and practical benchmarks for parallel-and-distributed processing. Less

我们开发了细粒度多线程处理器; Fuce架构用于并行和分布式处理。Fuce处理器以独占方式执行多个线程。Fuce的核心概念是基于延续的多线程执行，在本项目中，我们进行了以下研究：·开发了基于延续的计算模型，作为线程级并行计算的基础。Fuce-C是一种类C语言，它的编译系统支持基于连续的线程级并行处理。Fuce处理器在FPGA上实现。使用FPGA上的Fuce仿真器和软件仿真器评估了Fuce处理器的性能。·在基于延续的细粒度多线程技术的基础上设计了操作系统Cefos，并通过多个多线程处理基准测试评估了其可行性，验证了Fuce体系结构的有效性。我们实现了FUCE过程 ...更多信息在FPGA上用VHDL语言实现了仿真器，用Java语言实现了软件仿真器。利用FPGA仿真器和软件仿真器对素数筛选算法、FFT算法、快速排序算法、归并排序算法和N皇后问题等基准算法进行了测试。通过这种评估，我们表明，细粒度多线程是非常有效的各种并行计算：数据并行处理，流水线并行处理和它们的混合。特别是，基于连续的多线程技术是非常有用的高性能流处理，这是一个重要的处理方案，构建操作系统内核。作为评估的结果，我们得出结论，基于连续的细粒度多线程是一个非常有效的方法来解决并行处理问题，如操作系统内核的建设，需要基本的并发和并行处理技术。我们还评估了Fuce处理器的硬件成本，并表明Fuce处理器与商品处理器相比具有硬件成本效益。我们开发了基于延续的细粒度多线程程序的操作系统内核机制。我们还展示了使用基于延续的细粒度多线程技术可以简单地构建操作系统内核。研究了一种新的体系结构和软硬件协同工作的基本机制，即用于并行和分布式处理的Fuce内核，并证明了Fuce内核是开发实用并行处理器的可行性。本研究的目的是研究开发实用的并行和分布式机器的基本技术，现在我们可以得出结论，这一研究目的已经实现。我们将进一步努力使Fuce机器更加实用。我们在FPGS仿真器和软件仿真器上开发了Fuce-C语言系统和Cefos操作系统，并在商品集群机上运行。然后，我们将使用更可扩展和实用的并行和分布式处理基准来评估Fuce架构。少