A High-memory Supercomputer for Proteomics, Text Mining and Microbiome Research

描述(由申请人提供)：我们申请资金购买一台集成的超级计算机，将5个在微生物组、蛋白质组学、文本挖掘和超级计算方面具有互补专业知识的高效和协作实验室联合起来，并通过云和网络应用程序将这些能力扩展到更广泛的NIH资助的生物医学研究社区。园区中的其他系统无法满足的关键共享需求，商业云中无法提供的，以及国家实验室超额订阅的，是一个可以运行需要高内存(8-32 GB/核)和长时间(2周工作时间)的作业的系统，并且针对使其他系统上的网络或存储饱和的高IO任务进行了优化。该系统将由128台服务器组成，每台服务器使用2x8核2.93 GHz英特尔沙桥CPU。20个大内存节点每个有512 GB的RAM(32 GB/核)，100个计算节点每个有128 GB的RAM(8 GB/核)。这120个节点将使用两个绑定在一起的10Gbps以太网端口，以20Gbps/节点(2.5 Gb/S)的速度连接到系统的其余部分，每个节点将拥有2.4TB的原始高性能本地存储。这些本地磁盘的总聚合性能超过36 GB/S(&gt；300MB/S/节点)。其余8个节点将用于管理、支持高级软件工具和基础设施以及用户交互。一个中央高性能Lustre并行文件系统将为128个客户端提供1.15PB的可用暂存空间，并支持36 GB/S。一个由4个驱动器/300个磁带组成的归档系统将支持&gt；1 GB/S聚合(包括压缩)，提供450 TB的原始容量，存储约4.5 PB的用户数据，并扩展到此大小的5倍。该系统价值450万美元，但由于这一合作伙伴关系的战略重要性，惠普报价为200万美元，将安装在最先进的 Boulder校区新的Jennie Smoly Caruthers生物技术大楼的艺术机房(2012年2月开业)，并以40Gbps的速度连接到校园的其他部分。该系统将成为关键科学领域的关键支持技术，在这些领域，数据呈指数级增长，校园中的当前系统已报废，仅用于其他目的，或针对其他任务进行优化。主要用户将主要将该仪器用于耗时的一次性任务，如微生物组和基因组组装工作流程的参数优化、建立知识库、执行模拟和数据库搜索，这些资源将被更广泛的用户社区(数百名合作者；数千名最终用户)重新使用，这些用户社区没有超级计算访问权限。该提议的一个关键创新方面是将系统的一部分配置为学术云，这将使我们能够在商业云(例如Amazon EC2)和学术云(例如Magellan和DIAG)升级后，试验可由不同用户在这些云上部署的工作流。该系统还将通过向有前途的新教员和受训人员提供NIH培训补助金，以及与测序核心的新用户合作，在生命科学的高性能计算方面建立广泛的专业基础。因此，拟议的工具将对美国国立卫生研究院资助的研究产生深远影响。

DESCRIPTION (provided by applicant): We request funds to purchase an integrated supercomputer to unite 5 highly productive and collaborative laboratories with complementary expertise in the microbiome, proteomics, text mining, and supercomputing, and to extend these capabilities to the broader NIH-funded biomedical research community via cloud and web applications. The critical shared need not met by other systems on campus, unavailable in commercial clouds, and oversubscribed at national labs, is for a system that can run jobs that require high memory (8-32 GB/core) and long duration (>2 weeks wall-time), and is optimized for high-IO tasks that saturate network or storage on other systems. The system will consist of 128 servers, each using 2x8-core 2.93GHz Intel Sandybridge CPUs. 20 large-memory nodes will each have 512GB of RAM (32GB/core), and 100 compute nodes will each have 128GB of RAM (8GB/core). These 120 nodes will each use two 10Gbps Ethernet ports bonded together for a 20Gbps/node (2.5GB/s) connection to the rest of the system, and each node will have 2.4TB raw high- performance local storage. The total aggregate performance of these local disks is over 36GB/s sustained (>300MB/s per node). The remaining 8 nodes will be used for administration, support for advanced software tools and infrastructure, and user interaction. A central high-performance Lustre parallel file system will provide 1.15PB of usable scratch space and sustain 36GB/s to the 128 clients. An archival system of 4 drives/300 tapes will sustain >1GB/s aggregate (accounting for compression), provide 450TB of raw capacity, store ~4.5 PB of user data, and scale to 5x this size. The system, valued at $4.5 million but quoted at $2 million by HP due to the strategic importance of this partnership, will be housed in a state-of-the art machine room in the new Jennie Smoly Caruthers Biotechnology Building on the Boulder campus (opening Feb 2012), and connect to the rest of the campus at 40Gbps. The system will be a key enabling technology for key scientific areas where data growth is exponential and current systems on campus are end-of-life, solely dedicated to other purposes, or optimized for other tasks. The major users will use the instrument largely for time-consuming one-time tasks such as parameter optimization for microbiome and genome assembly workflows, building knowledgebases, and performing simulations and database searches that will provide resources that are re-used by much broader user communities (hundreds of collaborators; thousands of end users) who lack supercomputing access. One key innovative aspect of this proposal is configuration of part of the system as an academic cloud, which will allow us to pilot workflows that can later be deployed by diverse users on commercial clouds (e.g. Amazon EC2) and academic clouds (e.g. Magellan and DIAG) once those clouds are upgraded. The system will also build a broad expertise base in high-performance computing in the life sciences through outreach to promising new faculty and trainees on NIH training grants, and collaborations with new users of the Sequencing Core. The proposed instrument will thus have a profound impact on NIH-funded research.