Brain Science Computer Cluster

描述(由申请人提供)：当代脑科学研究的计算要求经常超过单个研究实验室的财务和资源管理限制。许多当代神经科学研究项目需要使用先进的统计方法和解剖重建技术来分析大数据集。这些方法需要在配备大容量高速存储设备的多处理器环境中运行的高速计算和图形引擎。布朗脑科学在理解神经处理方面的努力的一个限制是缺乏容易获得的高速计算资源。基于当代Linux CPU和GPU的统一集群的中央计算资源将服务于Brown的脑科学研究人员核心小组的计算需求，而不会影响独立工作站常见的个人访问。所请求的计算机集群具有自动平衡CPU和GPU使用率的系统软件，从而确保所有用户最大限度地使用计算资源。密集的3D图形被分流到GPU或客户端工作站，从而进一步减少了中央计算负荷。将使用具有开放操作环境的商业或开放源码软件，使用标准和新颖的统计和机器学习方法进行分析，以评估大型数据集的重要性。这项建议详细说明了当代计算资源的架构和好处，适用于主要和次要用户，更广泛地说，是布朗脑科学社区。该资源旨在满足布朗大脑科学家核心团队的即时和近期计算和存储需求。该系统可以根据需要随时进行扩展，无论是计算、存储还是新用户。现有核心调查组的扩展很容易发生，因为系统的计算能力或存储能力可以以相对较低的成本容易地增强。该系统的灵活性将服务于布朗脑科学界的各种研究需求。预计计算资源将把布朗大学致力于神经处理这一常见问题的研究人员聚集在一起。

DESCRIPTION (provided by applicant): Computational requirements of contemporary brain science research often exceed financial and resource management limits of individual investigator laboratories. Many contemporary neuroscience research projects require analysis of large data sets with advanced statistical methods and anatomical reconstruction techniques. These methods require high speed computational and graphics engines operating in a multiple processor environments equipped with large capacity, high-speed storage devices. A limitation in the Brown brain science effort at understanding neural processing is the lack of a readily accessible high-speed computational resource. A central computational resource based on a unified cluster of contemporary Linux CPUs and GPUs will serve the computational needs of a core group of brain science investigators at Brown without compromising individual access common to stand-alone workstations. The requested computer cluster has system software that automatically balances CPU and GPU usage, thereby ensuring maximum access to the computational resource for all users. Intensive 3D graphics are off-loaded either to GPUs or to client workstations, thereby further reducing the central computational load. Commercial or open-source software with an open operating environment will be used for analysis using standard and novel statistical and machine learning approaches to assess significance of large data sets. This proposal details the architecture and benefits of a contemporary computational resource for the major and minor users, and more generally the Brown brain science community. The resource was designed to fill immediate and near-term computational and storage needs of a core group of Brown brain scientists. The system can be readily expansion as needs, either computational, storage, or new users, arise. Expansion of the existing core investigators group can occur easily since the computational power or storage capacity of the system can be readily enhanced at relatively low cost. The flexible nature of the system will serve a variety of research needs of the Brown brain science community. The computational resource is expected to bring together researchers at Brown working on the common problem of neural processing.