MRI: Acquisition of SABER: Shared Analytics and Big-data Enterprise Resource

MRI：收购 SABRE：共享分析和大数据企业资源

• 批准号: 1626432
• 负责人: Farzad Mashayek
• 金额: $ 59.95万
• 依托单位: University of Illinois at Chicago
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-15 至 2019-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1626432&HistoricalAwards=false
• 关键词: MRI; Acquisition; SABER; Shared; Analytics

Computation continuous to grow across all scientific disciplines and research education. Along with computationally-driven research, vast amounts of data are originating from various sources such as particle colliders (e.g. Large Hadron Collider), environmental sensors, social media and transportation systems, etc. To handle this "Data Tsunami", Big-Data analysis is becoming pervasive across both academia and industry. To harness the power of these vast datasets and expedite scientific discovery through computation and analysis, institutions need adequate resources to respond to emerging opportunities. This public Minority-Serving Institution in the heart of the Chicago area, is the first to use an open access cluster in the region. Thus SABER should have a significant positive impact on research training for underrepresented groups. The institution?s unique approach to utilizing existing technologies should set the converging path for Big Data and HPC clusters nationally. SABER will greatly encourage Big Data research across various scientific disciplines such as engineering, natural sciences, urban planning, economics, manufacturing, and others at institutions across Chicago region. Advances made in algorithm development on parallel architectures and data mining techniques resulting from this acquisition should initiate transformative advances for all other types of computational research at the national level.This project, acquiring a hybrid HPC (High Performance Computing) and Big-Data cluster, aims to support more than 25 research projects from 9 colleges at UIC and IIT (Illinois Institute of Technology) in a broad range of computational disciplines such as fluid dynamics, physics, chemistry, biology, mathematics, bioengineering, visualization, earth and environmental sciences, and material sciences. This heterogeneous cluster, SABER (Shared Analytics and Big-data Enterprise Resource) named SABER, has diskless nodes allowing on-demand research cloud provisioning services, fat local storage nodes for MapReduce/Big-Data analysis, petascale high capacity fast-scratch storage and large memory nodes for High Throughput Computing, all connected over 56Gbps FDR Infiniband. SABER complements UIC's existing private condo (for investors only) cluster, EXTREME, by bridging the gap between available infrastructure and desired capabilities, such as Big Data analysis, dual operating systems capability and a fee-for-service model. SABER will be UIC's and IIT's first open access cluster (this instrument does not currently exist at either institution), which will also connect to Internet2 and Pacific Research Platform (PRP) over 100Gbps, thus facilitating multi-site and multi-investigator research involving ultra-large datasets. The primary use of SABER will not be for research with disease-related goals. Traditionally silo clusters have been created for HPC and Big Data type workloads. The SABER cluster will employ novel technical advances made in the fast scratch Lustre environment by Intel to handle both HPC and Big Data workloads on the same hardware, thus leading to greater economies of scale. SABER connects to UIC's Electronic Visualization Laboratory over a high-speed 100Gbps channel to enable live visualization of real time computed data at remote locations. SABER?s shared storage architecture allows researchers and students to compute, analyze and visualize massive datasets without costly data transfers. In addition, researchers will be able to cloud provision Windows and Linux mini-clusters at run times that break the traditional environmental constraints of HPC Linux, to allow greater adoption.

计算在所有科学学科和研究教育中不断增长。伴随着计算驱动的研究，海量数据来自各种来源，如粒子对撞机(例如大型强子对撞机)、环境传感器、社交媒体和交通系统等。为了应对这场“数据海啸”，大数据分析正在学术界和工业界普及。为了利用这些海量数据集的力量，并通过计算和分析加速科学发现，各机构需要足够的资源来应对新出现的机会。这所公共少数群体服务机构位于芝加哥地区的中心，是该地区第一个使用开放访问集群的机构。因此，SABER应该对代表性不足群体的研究培训产生重大的积极影响。S利用现有技术的独特方法应该为全国范围内的大数据和高性能计算集群设定融合之路。Saber将极大地鼓励芝加哥地区各机构进行跨不同科学学科的大数据研究，如工程、自然科学、城市规划、经济学、制造业和其他学科。此次收购在并行体系结构和数据挖掘技术的算法开发方面取得的进展，应该会为国家层面的所有其他类型的计算研究带来变革性的进展。该项目收购了一个混合的HPC(高性能计算)和大数据集群，旨在支持来自UIC和IIT(伊利诺伊理工学院)9所学院的25个以上研究项目，涉及广泛的计算学科，如流体动力学、物理、化学、生物、数学、生物工程、可视化、地球和环境科学以及材料科学。这个名为SABER(共享分析和大数据企业资源)的异类群集拥有支持按需研究云资源调配服务的无盘节点、用于MapReduce/大数据分析的胖本地存储节点、用于高吞吐量计算的千万亿级高容量快速擦除存储和用于高吞吐量计算的大内存节点，所有这些节点都通过56 Gbps FDR Infiniband连接。Saber通过弥合现有基础设施和所需功能之间的差距，如大数据分析、双操作系统功能和收费服务模式，补充了UIC现有的私人公寓(仅限投资者)集群Extreme。SABER将是UIC和IIT的第一个开放获取集群(这两个机构目前都不存在这种仪器)，它还将以100Gbps的速度连接到Internet2和太平洋研究平台(PRP)，从而促进涉及超大数据集的多站点和多研究者研究。军刀的主要用途不是用于与疾病相关的研究。传统上，思洛群集是为HPC和大数据类型工作负载创建的。Saber群集将采用英特尔在FAST Scratch Lustre环境中取得的新技术进步，在同一硬件上处理HPC和大数据工作负载，从而带来更大的规模经济。Saber通过高速100Gbps通道连接到UIC的电子可视化实验室，以实现远程实时计算数据的实时可视化。Saber-S共享存储架构允许研究人员和学生计算、分析和可视化海量数据集，而无需昂贵的数据传输。此外，研究人员将能够在运行时对Windows和Linux迷你集群进行云配置，打破了HPC Linux的传统环境限制，从而允许更多的人采用。