Project Proposed: This project, acquiring a hybrid high-performance GPU (graphics processing unit)/CPU system, enables broader heterogeneous computing by deploying multiple types of computing nodes and allowing each to perform the tasks to which it is best suited in traditional CPU-based, GPU based, and hybrid GPU/CPU applications. Satisfying two major research environments in computer and information sciences and scientific computing, the instrument - Serves faculty and students conducting research using existing parallel application software and developing tools for parallel programming and parallel applications of the system, - Serves the center for High Performance Computing and Networking at the institution and the greater Philadelphia region offering services not only to other departments on campus, but also to other institutions in the local community, such as area high schools and local colleges/universities , - Educates and trains providing a computing environment to various science courses for students to gain hands-on experience and offers research training sessions to the local research community; and - Fosters collaboration by supporting joint research with local colleges/universities, in collaboration with other schools in the state, to develop, test, and apply advanced tools for designing and executing parallel programs. Among others, the Center services Chemistry, Computer and Information Sciences, Electrical and Computer Engineering, Mathematics, Physics, and Pulmonary, Critical Care Medicine and Physical Therapy. The instrument specifically supports research projects with broad impact in molecular self assembly, microvessel networks, spatio-temporal data analysis, large-scale system simulation, effective uniformization, fault tolerant computing, etc. and augments existing applications using the GPU as an accelerator enabling some problems to run entirely on the GPU. Broader Impacts: The instrument greatly enhances the current computing facilities at the institution. With its GPU/CPU component, this instrument is the first of its kind of high-performance computing facility in the greater Philadelphia region, an area with a high degree of diversity. This university draws a substantial portion of its students from one of the largest African American populations in the country. Enabling education, training, and collaboration, the instrument contributes to foster economic growth in an area with a high concentration of high-tech and IT-related industries that currently has no high-performance computing and networking center of this magnitude. The region is now able to go beyond minimal services and promote and support collaboration and cooperation across sectors (e.g., higher education and local industry).
Major Activities: In this reporting period, we have conducted the following major research activities: (1) Built and launched Owlsnest HPC cluster for the Temple University research and education communities. (2) Built and launched TCloud private HPC cloud for the Temple University research and education communities. (3) Conducted computational experiments in Computer Sciences, Chemistry, Physics, Biology, Mathematics and Engineering (4) Developed and taught both undergraduate and graduate mobile, cloud and parallel processing courses since the production of MRI facilities. (5) Built a secure Cloud storage and processing scheme to can keep user privacy while maintain system flexibility. Specific Objectives: (a) To enable Temple faculty and students have direct access to often talked-about HPC clusters and tools. (b) To foster cross-disciplinary computational research and developments. (c) To develop, experiment and compare different HPC provisioning methods using the traditional HPC resources (Owlsnest vs. TCloud). (d) To conduct both compute intensive and data intensive research and development projects. Significant Results: Our significant results are the following: (1) Computational Chemistry (Klein):[ Significant GPU acceleration of traditional multicore molecular simulation packages using HOOMD]; a widely used database of "hypothetical zeolites" in the crystallographic community. (2) Mathematics (Rivin): [Geometry of groups, random knots and random polynomials]; (3) Electrical and Computer Engineering (Biswas): [Optimization and state estimation of large scale heterogeneous systems ]; (4) Computer and Network Sciences (Wu): [Optimization and scheduling of Heterogeneous Computing Systems ]; (5) Computing Architectures (Shi): [Volatility Harnessing Architectures for compute intensive and data intensive applications]; (6) Data Intensive Analytics (Obtradovic): [a novel approach for simultaneous links and attributes prediction in the temporal social networks, a conditional random fields probabilistic model for structured regression that uses multiple non-structured predictors as its features, a distributed clinical decision support tool that facilitates knowledge building using statistics based on patient data from multiple sites that satisfy a certain filtering condition without the need for actual data to leave the participating sites, an iterative method that alternately provides the maximum a posterior (MAP) estimation of the prediction and the maximum-likelihood (ML) estimation of quality of multiple predictors, Knowledge Discovery in High-Dimensional Spatio-Temporal Databases] The collaborative research has also resulted in a series of hands-on training workshops, local and international conferences: NSF/TCPP Curriculum on Parallel/Distributed Computing (Jie Wu) Introduction to Parallel Programming x 2(Axel Koymeyer) Introduction to TCloud (Justin Shi) Introductoon to Python and MPI (Rashad) University-wide CROO conference (Math, Chemistry, Biology, Physics, Electrical and Computer Engineering and Computer Information Sciences) Temple University Big Data Conference International Workshop on Sustainable HPC Cloud 2010-2013 (Banff, CA, Liverpool, UK, Salt Lake City, Utah, Denver, CO) (Justin Shi) International Symposium on InterCloud HPC (HPDC2013, Helsinki, Finland)(Justin Shi) In addition to campus-wide training and research exchanges, faculty and students involved in computational sciences are also actively involved in international workshops, conferences and journal publications. We have also developed mobile, cloud and HPC curriculums for undergraduate and graduate students. These include Graduate class on "Ad Hoc Networks" in Spring 2013 (Jie Wu) Graduate class on "Distributed and Parallel Computing in Spring 2013 (Justin Shi) Graduate class on "Mobile and HPC Cloud Programming" in Fall 2011 (Chiu Tan) Undergraduate class on "Mobile Programming Technologies in Fall 2010-2013 (Justin Shi) Several NSF REU site students at Temple University are involved in computational science research in summers of 2012 and 2013. Faculty and graduate students have also contributed to the Open Source HPC solutions. These include CUDA programming solutions for HOOMD by Axel Kolmeyer Automatic HPC Cluster configuration tool HPCFY by Moussa Taifi Statistic Multiplexed HPC Computing Toolkit SynergyV3.0+ by Justin Y. Shi (jys673@github.com) The Open Source solutions are hosted at github.com. The students on these projects are trained on a variety of cross-cutting disciplines. These include network optimization, probabilistic analysis, approximation and low complexity algorithms, GPGPU solutions, HPC cloud and traditional Cluster comparative research, extreme scale fault tolerance software development, and experimentation on real systems. This multi-disciplinary training helps them be better prepared for both industry and academia.
