/pdf/scalable-molecular-dynamics-with-namd-3j7xyc70g4.pdf

Scalable Molecular Dynamics with NAMD

Author(s): Asanovic, K; Bodik, R; Catanzaro, B; Gebis, J; Husbands, P; Keutzer, K; Patterson, D; Plishker, W; Shalf, J; Williams, SW | Abstract: The recent switch to parallel microprocessors is a milestone in the history of computing. Industry has laid out a roadmap for multicore designs that preserves the programming paradigm of the past via binary compatibility and cache coherence. Conventional wisdom is now to double the number of cores on a chip with each silicon generation. A multidisciplinary group of Berkeley researchers met nearly two years to discuss this change. Our view is that this evolutionary approach to parallel hardware and software may work from 2 or 8 processor systems, but is likely to face diminishing returns as 16 and 32 processor systems are realized, just as returns fell with greater instruction-level parallelism. We believe that much can be learned by examining the success of parallelism at the extremes of the computing spectrum, namely embedded computing and high performance computing. This led us to frame the parallel landscape with seven questions, and to recommend the following: • The overarching goal should be to make it easy to write programs that execute efficiently on highly parallel computing systems • The target should be 1000s of cores per chip, as these chips are built from processing elements that are the most efficient in MIPS (Million Instructions per Second) per watt, MIPS per area of silicon, and MIPS per development dollar. • Instead of traditional benchmarks, use 13 “Dwarfs” to design and evaluate parallel programming models and architectures. (A dwarf is an algorithmic method that captures a pattern of computation and communication.) • “Autotuners” should play a larger role than conventional compilers in translating parallel programs. • To maximize programmer productivity, future programming models must be more human-centric than the conventional focus on hardware or applications. • To be successful, programming models should be independent of the number of processors. • To maximize application efficiency, programming models should support a wide range of data types and successful models of parallelism: task-level parallelism, word-level parallelism, and bit-level parallelism. 1 The Landscape of Parallel Computing Research: A View From Berkeley • Architects should not include features that significantly affect performance or energy if programmers cannot accurately measure their impact via performance counters and energy counters. • Traditional operating systems will be deconstructed and operating system functionality will be orchestrated using libraries and virtual machines. • To explore the design space rapidly, use system emulators based on Field Programmable Gate Arrays (FPGAs) that are highly scalable and low cost. Since real world applications are naturally parallel and hardware is naturally parallel, what we need is a programming model, system software, and a supporting architecture that are naturally parallel. Researchers have the rare opportunity to re-invent these cornerstones of computing, provided they simplify the efficient programming of highly parallel systems.

/pdf/the-landscape-of-parallel-computing-research-a-view-from-1i5pswggnf.pdf

The Landscape of Parallel Computing Research: A View from Berkeley

ABSTRA C T We simulate the assembly of a massive rich cluster and the formation of its constituent galaxies in a flat, low-density universe. Our most accurate model follows the collapse, the star formation history and the orbital motion of all galaxies more luminous than the Fornax dwarf spheroidal, while dark halo structure is tracked consistently throughout the cluster for all galaxies more luminous than the SMC. Within its virial radius this model contains about 2 10 7 dark matter particles and almost 5000 distinct dynamically resolved galaxies. Simulations of this same cluster at a variety of resolutions allow us to check explicitly for numerical convergence both of the dark matter structures produced by our new parallel N-body and substructure identification codes, and of the galaxy populations produced by the phenomenological models we use to follow cooling, star formation, feedback and stellar aging. This baryonic modelling is tuned so that our simulations reproduce the observed properties of isolated spirals outside clusters. Without further parameter adjustment our simulations then produce a luminosity function, a mass-to-light ratio, luminosity, number and velocity dispersion profiles, and a morphology ‐radius relation which are similar to those observed in real clusters. In particular, since our simulations follow galaxy merging explicitly, we can demonstrate that it accounts quantitatively for the observed cluster population of bulges and elliptical galaxies.

Populating a cluster of galaxies - I. Results at z=0

SUMMARY Since 1984, the Condor project has enabled ordinary users to do extraordinary computing. Today, the project continues to explore the social and technical problems of cooperative computing on scales ranging from the desktop to the world-wide computational Grid. In this paper, we provide the history and philosophy of the Condor project and describe how it has interacted with other projects and evolved along with the field of distributed computing. We outline the core components of the Condor system and describe how the technology of computing must correspond to social structures. Throughout, we reflect on the lessons of experience and chart the course travelled by research ideas as they grow into production systems. Copyright c � 2005 John Wiley & Sons, Ltd.

/pdf/distributed-computing-in-practice-the-condor-experience-1xldnwf3zo.pdf

Distributed computing in practice: the Condor experience

Ganglia is a scalable distributed monitoring system for high performance computing systems such as clusters and Grids. It is based on a hierarchical design targeted at federations of clusters. It relies on a multicast-based listen/announce protocol to monitor state within clusters and uses a tree of point-to-point connections amongst representative cluster nodes to federate clusters and aggregate their state. It leverages widely used technologies such as XML for data representation, XDR for compact, portable data transport, and RRDtool for data storage and visualization. It uses carefully engineered data structures and algorithms to achieve very low per-node overheads and high concurrency. The implementation is robust, has been ported to an extensive set of operating systems and processor architectures, and is currently in use on over 500 clusters around the world. This paper presents the design, implementation, and evaluation of Ganglia along with experience gained through real world deployments on systems of widely varying scale, configurations, and target application domains over the last two and a half years.

/pdf/the-ganglia-distributed-monitoring-system-design-3h3k67tte9.pdf

The ganglia distributed monitoring system: design, implementation, and experience

Accelerators, most notably graphics processing units (GPUs), are an increasingly common component of large computing installations due to their high peak performance and competitive power efficiency figures. However, because of the substantially different architecture, their programming requires an intimate knowledge of the internal organization of specific devices, complicating their integration as a system component of equal utility to that of conventional CPUs and reducing code portability. The evolution of GPU programming toolkits and libraries such as Nvidia's Compute Unified Device Architecture has made this task easier, but has not fully eliminated a significant learning curve and barrier to application. In contrast, OpenACC is considered to be one of the easiest accelerator programming environments to master, albeit at the cost of lower computational efficiency for certain problem types. It leverages the same directive-based approach as OpenMP, with which it also shares many keywords and concepts.

The Essential OpenACC

http://ieeexplore.ieee.org/iel1/2/11956/00546609.pdf

Improving Performance on Exemplar

Control parallelism and data parallelism is mostly reasoned and optimized as separate functions. Because of this, workloads that are irregular, fine-grain and dynamic such as dynamic graph processing become very hard to scale. An experimental research approach to computer architecture that synthesizes prior techniques of parallel computing along with new innovations is proposed in this paper. We establish the background and motivation of the research undertaking and provide a detailed description of the proposed omputing system that is highly parallel non-von Neumann, memory-centric and memory-driven. We also present a message-driven (or even-driven) programming model called"diffusive computation"and provide insights into its properties using SSSP and Triangle Counting problems as examples.

Uniting Control and Data Parallelism: Towards Scalable Memory-Driven Dynamic Graph Processing

The emergence of new classes of HPC systems where performance improvement is enabled by Moore’s Law for technology is manifest through multi-core-based architectures including specialized GPU structures. Operating systems were originally designed for control of uniprocessor systems. By the 1980s multiprogramming, virtual memory, and network interconnection were integral services incorporated as part of most modern computers. HPC operating systems were primarily derivatives of the Unix model with Linux dominating the Top-500 list. The use of Linux for commodity clusters was first pioneered by the NASA Beowulf Project. However, the rapid increase in number of cores to achieve performance gain through technology advances has exposed the limitations of POSIX general-purpose operating systems in scaling and efficiency. This project was undertaken through the leadership of Sandia National Laboratories and in partnership of the University of New Mexico to investigate the alternative of composable lightweight kernels on scalable HPC architectures to achieve superior performance for a wide range of applications. The use of composable operating systems is intended to provide a minimalist set of services specifically required by a given application to preclude overheads and operational uncertainties (“OS noise”) that have been demonstrated to degrade efficiency and operational consistency. This project was undertakenmore » as an exploration to investigate possible strategies and methods for composable lightweight kernel operating systems towards support for extreme scale systems.« less

/pdf/a-framework-for-adaptable-operating-and-runtime-systems-3r3xbalvt1.pdf

A Framework for Adaptable Operating and Runtime Systems

Although distance learning has a history that spans many decades, the full opportunity that is implicit in its exploitation has not been fully realized due to combination of factors including disparate experience between it and its classroom counterpart. However current and emerging technologies are helping overcome this barrier by providing significantly better interaction among the individual participants, thereby opening new avenues for knowledge dissemination. LSU in collaboration with five other institutions has developed effective methods that greatly extend the educational opportunities through combination of advanced technologies and educational methodologies. LSU and its partners have tested these technologies in realtime over the last two years. While further improvements are needed, this activity represents the current state of the art in technologies utilized and the quality of content and experience delivered. The distance learning initiative undertaken by LSU and its partners is driven by a vision for education, which aims to deliver expert & top-quality educational content to locations irrespective of their economic or technological

Thomas Sterling

Papers

The Essential OpenACC

Improving Performance on Exemplar

Uniting Control and Data Parallelism: Towards Scalable Memory-Driven Dynamic Graph Processing

A Framework for Adaptable Operating and Runtime Systems

Distance learning with high-definition low latency video and multi-media infrastructure