CAREER: Multicore-Based Parallel Disk Systems for Large-Scale Data-Intensive Computing

为了提高磁盘I/O性能，可以将多核处理器与并行磁盘系统集成在一起。不幸的是，多核并行磁盘系统的体系结构和数据处理算法仍处于起步阶段。这促使我们为并行磁盘系统开发新的体系结构，其中重要的多核处理能力和内存集成到并行磁盘驱动器中。这个CAREER项目提供了第一个并行磁盘系统，其中大部分数据和I/O处理被卸载到嵌入磁盘驱动器的多核处理器上。所建议的技术和机制对具有大小磁盘请求的动态工作负载具有高度的适应性，使现代并行磁盘系统利用多核处理器来获得更好的性能和可伸缩性。这个CAREER Development项目的总体目标是构建硬件和软件并行磁盘架构，将大量的多核计算能力放在磁盘上。本研究包括三个基本任务：(1)设计基于多核并行磁盘系统的软硬件架构；(2)开发基于多核的数据处理技术；(3)构建软件性能模型和分析工具包。教育计划包括：(1)建立存储系统实验室；（二）开设新课程；(3)实施小会议教育理念；(4)增加代表性不足的学生参与研究活动。此外，该项目将通过为下一代并行磁盘系统开发硬件和软件模块来造福社会，其中多核处理器和磁盘驱动器紧密集成以提高磁盘I/O性能。

To improve disk I/O performance, one may integrate multicore processors with parallel disk systems. Unfortunately, architectures and data processing algorithms for multicore-based parallel disk systems are still in their infancy. This motivates us to develop novel architectures for parallel disk systems, where significant multicore processing power and memory are integrated into parallel disk drives. This CAREER project provides the first parallel disk system in which large parts of data and I/O processing are offloaded to multicore processors, embedded in disk drives. The proposed techniques and mechanisms are highly adaptive to dynamic workloads with both large and small disk requests, making modern parallel disk systems leverage multicore processors for better performance and scalability.The overall objective of this CAREER Development project is to build hardware and software parallel disk architectures that put substantial multicore computing power on disks. The research consists of three basic tasks: (1) designing hardware and software architectures for multicore-based parallel disk systems, (2) developing multicore-based data processing techniques, and (3) building software performance models and an analysis toolkit. The educational plan include: (1) establishing a storage systems laboratory; (2) developing new courses; (3) implementing the concept of a mini-conference to educate students; and (4) increasing underrepresented student involvement in research activities. In addition, the project would benefit society by developing hardware and software modules for next-generation parallel disk systems, where multicore processors and disk drives are tightly integrated to boost disk I/O performance.