MRI: Development of a GPU-Enabled, Petascale Active Storage Architecture for Data-Intensive Applications in HPC and Cloud Environments

提案#：12-29282PI(S)：Skjellum，Anthony banalore，Purushotham；Hasan，Ragib；张成翠研究所：阿拉巴马大学伯明翰分校标题：MRI/Dev.：一种支持GPU的Petascale主动存储架构，适用于高性能计算和云环境中的数据密集型应用建议：该项目开发2.4PB的原始存储工具，以支持各种实验性HPC和云存储的研究项目，旨在为科学计算增加本地资源，并作为支持GPU的可靠存储的试验台。该工具支持更高的存储虚拟化、对故障情况下的存储的并发访问(例如，RAID)以及一系列数据密集型应用程序。将利用现有系统吸取的经验教训，将现有系统和新系统整合在一起，以支持云计算和灾后恢复的运作模式。该项目使以下研究和研究项目成为可能：--在高度精炼的水平上研究有效的错误率和可靠性，并寻求识别和管理更多类别错误(例如，误导写入)的方法；--创建半分析模型，以便在寿命--可靠性--性能成本空间内实现可调的存储特性；--运行来自数据挖掘的应用程序(包括生物信息学，作为证明最终系统有效性的驱动力)，以便在这些数据密集型领域实现新的科学；和-进行计算机科学研究，旨在简化活动存储计算的使用。广泛影响：此仪器增加了机构-S在面向数据密集型应用程序的廉价、快速、实用、可靠的PB级存储中进行尖端研究的能力。强大的计算能力在逻辑上接近那个存储空间，这使得新的科学成为可能。将强调学生培训(包括代表人数不足的群体)。通过这一努力传播知识可能意义重大。

Proposal #: 12-29282PI(s): Skjellum, Anthony Bangalore, Purushotham; Hasan, Ragib; Zhang, ChengcuiInstitution: University of Alabama at BirminghamTitle: MRI/Dev.: A GPU-Enabled, Petascale Active Storage Architecture for Data-Intensive Applications in HPC and Cloud EnvironmentsProject Proposed:This project, developing a 2.4 Petabytes (PB) of raw storage instrument to support a variety of research projects in experimental HPC and cloud storage, aims to both increase local resources for scientific computing and act as a testbed for GPU-enabled reliable storage. The instrument enables an increased virtualization of storage, the concurrent access to storage under fault scenarios (e.g., RAID), and a series of data intensive applications. Lessons learned will be leveraged from the existing system in place, whereby the existing system and the new system will be integrated in a way that supports cloud and disaster recovery modes of operation. The project enables the following studies and research projects: - Studying of effective rates of errors and reliability at highly refined levels and seeking means to identify and manage additional classes of errors (e.g., misdirected writes);- Creating semi-analytical models to allow tunable storage characteristics within a lifetime-reliability-performance cost space; - Running applications from data mining (including bioinformatics as drivers for proving the efficacy of the final system), to achieve new science in these data-intensive domains; and- Conducting computer science research aimed at simplifying use of active storage computation.Broader Impacts: This instrumentation increases the institution?s capacity to conduct cutting-edge research in an inexpensive, fast, practical, reliable petascale storage for data-intensive applications. Significant computational power logically close to that storage enables new science. Student training (including underrepresented groups) will be emphasized. The knowledge dissemination through this effort could be significant.