High End Computing

For parallel applications running on high-end computing systems, which processes of an application get launched on which processing cores is typically determined at application launch time without any information about the application characteristics. As high-end computing systems continue to grow in scale, however, this approach is becoming increasingly infeasible for achieving the best performance. For example, for systems such as IBM Blue Gene and Cray XT that rely on flat 3D torus networks, process communication often involves network sharing, even for highly scalable applications. This causes the overall application performance to depend heavily on how processes are mapped on the network. In this paper, we first analyze the impact of different process mappings on application performance on a massive Blue Gene/P system. Then, we match this analysis with application communication patterns that we allow applications to describe prior to being launched. The underlying process management system can use this combined information in conjunction with the hardware characteristics of the system to determine the best mapping for the application. Our experiments study the performance of different communication patterns, including 2D and 3D nearest-neighbor communication and structured Cartesian grid communication. Our studies, that scale up to 131,072 cores of the largest BG/P system in the United States (using 80% of the total system size), demonstrate that different process mappings can show significant difference in overall performance, especially on scale. For example, we show that this difference can be as much as 30% for P3DFFT and up to twofold for HALO. Through our proposed model, however, such differences in performance can be avoided so that the best possible performance is always achieved.

The use of High Performance Computing (HPC) in commercial and consumer IT applications is becoming popular. HPC users need the ability to gain rapid and scalable access to high-end computing capabilities. Cloud computing promises to deliver such a computing infrastructure using data centers so that HPC users can access applications and data from a Cloud anywhere in the world on demand and pay based on what they use. However, the growing demand drastically increases the energy consumption of data centers, which has become a critical issue. High energy consumption not only translates to high energy cost which will reduce the profit margin of Cloud providers, but also high carbon emissions which are not environmentally sustainable. Hence, there is an urgent need for energy-efficient solutions that can address the high increase in the energy consumption from the perspective of not only the Cloud provider, but also from the environment. To address this issue, we propose near-optimal scheduling policies that exploit heterogeneity across multiple data centers for a Cloud provider. We consider a number of energy efficiency factors (such as energy cost, carbon emission rate, workload, and CPU power efficiency) which change across different data centers depending on their location, architectural design, and management system. Our carbon/energy based scheduling policies are able to achieve on average up to 25% of energy savings in comparison to profit based scheduling policies leading to higher profit and less carbon emissions.

In this paper, we present PyPANCG, a Python library-interface that implements both the conjugate gradient method and the preconditioned conjugate gradient method for solving nonlinear systems. We describe the use of the library and its advantages in order to get fast development. The aim of this library is to develop high performance scientific codes for high-end computers hiding many of the underlying low-level programming complexities from users with the use of a high-level Python interface. The library has been designed for adapting to different stages of the design process, depending on whether the purpose is computational performance or fast development. Experimental results report the performance of our approach on different parallel computers.

The rapid growth of InfiniBand, 10 Gigabit Ethernet/iWARP and IB WAN extensions is increasingly gaining momentum for designing high end computing clusters and data-centers. For typical applications such as data staging, content replication and remote site backup etc., FTP has been the most popular method to transfer bulk data within and across these clusters or data-centers. Although the existing sockets based FTP approaches can be transparently used in these systems through the protocols like IPoIB or SDP, their performance and scalability are limited due to the additional interaction overhead and unoptimized protocol processing. This leads to a challenge how to design more efficient FTP mechanisms by leveraging the advanced features of modern interconnects.

The use of High Performance Computing (HPC) in commercial and consumer IT applications is becoming popular. They need the ability to gain rapid and scalable access to high-end computing capabilities. Cloud computing promises to deliver such a computing infrastructure using data centers so that HPC users can access applications and data from a Cloud anywhere in the world on demand and pay based on what they use. However, the growing demand drastically increases the energy consumption of data centers, which has become a critical issue. High energy consumption not only translates to high energy cost, which will reduce the profit margin of Cloud providers, but also high carbon emissions which is not environmentally sustainable. Hence, energy-efficient solutions are required that can address the high increase in the energy consumption from the perspective of not only Cloud provider but also from the environment. To address this issue we propose near-optimal scheduling policies that exploits heterogeneity across multiple data centers for a Cloud provider. We consider a number of energy efficiency factors such as energy cost, carbon emission rate, workload, and CPU power efficiency which changes across different data center depending on their location, architectural design, and management system. Our carbon/energy based scheduling policies are able to achieve on average up to 30% of energy savings in comparison to profit based scheduling policies leading to higher profit and less carbon emissions.

Software optimization for multicore architectures is one of the most critical challenges in today's high-end computing. In this paper we focus on a well-known multicore platform, namely the Cell BE processor, and we address the problem of allocating and scheduling its processors, communication channels and memories, with the goal of minimizing application execution time. We have developed a complete optimization strategy based on Benders' decomposition. Unfortunately, a traditional two-stage decomposition produces unbalanced components: the allocation part is difficult, while the scheduling part is much easier. To address this issue, we have developed a multi-stage decomposition, which is a recursive application of standard Logic based Benders' Decomposition (LBD). Our experiments demonstrate that this approach is very effective in obtaining balanced sub-problems and in reducing the runtime of the optimizer.

This paper addresses the underlying sources of performance degradation (e.g. latency, overhead, and starvation) and the difficulties of programmer productivity (e.g. explicit locality management and scheduling, performance tuning, fragmented memory, and synchronous global barriers) to dramatically enhance the broad effectiveness of parallel processing for high end computing. We are developing a hierarchical threaded virtual machine (HTVM) that defines a dynamic, multithreaded execution model and programming model, providing an architecture abstraction for HEC system software and tools development. We are working on a prototype language, LITL-X (pronounced "little-X") for Latency Intrinsic-Tolerant Language, which provides the application programmers with a powerful set of semantic constructs to organize parallel computations in a way that hides/manages latency and limits the effects of overhead. This is quite different from locality management, although the intent of both strategies is to minimize the effect of latency on the efficiency of computation. We will work on a dynamic compilation and runtime model to achieve efficient LITL-X program execution. Several adaptive optimizations will be studied. A methodology of incorporating domainspecific knowledge in program optimization will be studied. Finally, we plan to implement our method in an experimental testbed for a HEC architecture and perform a qualitative and quantitative evaluation on selected applications.

The slowing pace of commodity microprocessor performance improvements combined with ever-increasing chip power demands has become of utmost concern to computational scientists. As a result, the high performance computing community is examining alternative architectures that address the limitations of modern cache-based designs. In this work, we examine the potential of using the forthcoming STI Cell processor as a building block for future high-end computing systems. Our work contains several novel contributions. First, we introduce a performance model for Cell and apply it to several key scientific computing kernels: dense matrix multiply, sparse matrix vector multiply, stencil computations, and 1D/2D FFTs. The difficulty of programming Cell, which requires assembly level intrinsics for the best performance, makes this model useful as an initial step in algorithm design and evaluation. Next, we validate the accuracy of our model by comparing results against published hardware results, as well as our own implementations on the Cell full system simulator. Additionally, we compare Cell performance to benchmarks run on leading superscalar (AMD Opteron), VLIW (Intel Itanium2), and vector (Cray X1E) architectures. Our work also explores several different mappings of the kernels and demonstrates a simple and effective programming model for Cell's unique architecture. Finally, we propose modest microarchitectural modifications that could significantly increase the efficiency of double-precision calculations. Overall results demonstrate the tremendous potential of the Cell architecture for scientific computations in terms of both raw performance and power efficiency.

In this work we present an scientific application that has been given a Hadoop MapReduce implementation. We also discuss other scientific fields of supercomputing that could benefit from a MapReduce implementation. We recognize in this work that Hadoop has potential benefit for more applications than simply datamining, but that it is not a panacea for all data intensive applications.

7 MW [l]. There is a critical need for new hardware technologies as well as new architectural approaches to Scaling CMOS to higher pe$ormance for high end achieve these performance levels at acceptable power. computing requires limiting prime power and improving The Hybrid Technology, Multi-Threaded (HTMT) computing efficiency. In space, rad-hard Parts are architecture study [ 2 ] identified advanced hardware and circuitry is an attractive solution. We project 50 Kgates/ch@ operating with 100 G H~ clock while incremental advances. We present here a revolutionary dissipating only -5 m W O Total power per chip, including Processor hardware approach for high end Computingcooling to 5 K, is -2.5 W SFQ processor chips will be superconductor single-flux-quantum (SFQ) LSI logic. more efficient than CMOS. In addition, > 99 % of the power is unregulated. Extremely low supply voltage, as 1 1 E~~~,~~~~ low as 2 pV/GHz, enables the ultra-low power of SFQ circuits. A 100 GHz clock requires only 200 pV. FLUX is a technology development program to validate SFQ Total power is a critical factor for both ground-and technology for high end computing: LSI fabrication, chip space-based computing. State-of-the-art supercomputers architecture, design methodologies, 50 Gb/s inter-chip require large footprints and megawatts of power, limiting communication, and packaging. We discuss the status siting o p p o~t i e s . he can anticipate that this situation processors for a petaFLOPS machine. The baseline Hybrid Technology, Multi-Threaded design uses 4096 level. SFR processors operating at 50-100 GHz to achieve 10 FLOPS. Total SFQprocessorpower is -250 kW required.

Adaptive mesh refinement (AMR) is a powerful technique that reduces the resources necessary to solve otherwise in-tractable problems in computational science. The AMR strategy solves the problem on a relatively coarse grid, and dynamically refines it in regions requiring higher resolution. However, AMR codes tend to be far more complicated than their uniform grid counterparts due to the software infrastructure necessary to dynamically manage the hierarchical grid framework. Despite this complexity, it is generally believed that future multi-scale applications will increasingly rely on adaptive methods to study problems at unprecedented scale and resolution. Recently, a new generation of parallel-vector architectures have become available that promise to achieve extremely high sustained performance for a wide range of applications, and are the foundation of many leadership-class computing systems worldwide. It is therefore imperative to understand the tradeoffs between conventional scalar and parallel-vector platforms for solving AMR-based calculations. In this paper, we examine the HyperCLaw AMR framework to compare and contrast performance on the Cray X1E, IBM Power3 and Power5, and SGI Altix. To the best of our knowledge, this is the first work that investigates and characterizes the performance of an AMR calculation on modern parallel-vector systems.

With the latest high-end computing nodes combining shared-memory multiprocessing with hardware multithreading, new scheduling policies are necessary for workloads consisting of multithreaded applications. The use of hybrid multiprocessors presents schedulers with the problem of job pairing, ie deciding which specific jobs can share each processor with minimum performance penalty, by running on different execution contexts. Therefore, scheduling policies are expected to decide not only which job mix will ...

Many scientific programs exchange large quantities of double-precision data between processing nodes and with mass storage devices. Data compression can reduce the number of bytes that need to be transferred and stored. However, data compression is only likely to be employed in high-end computing environments if it does not impede the throughput. This paper describes and evaluates FPC, a fast lossless compression algorithm for linear streams of 64-bit floating-point data. FPC works well on hard-to-compress scientific datasets and meets the throughput demands of high-performance systems. A comparison with five lossless compression schemes, BZIP2, DFCM, FSD, GZIP, and PLMI, on four architectures and thirteen datasets shows that FPC compresses and decompresses one to two orders of magnitude faster than the other algorithms at the same geometric-mean compression ratio. Moreover, FPC provides a guaranteed throughput as long as the prediction tables fit into the L1 data cache. For example, on a 1.6 GHz Itanium 2 server, the throughput is 670 megabytes per second regardless of what data are being compressed.

As High-End Computing machines continue to grow in size, issues such as fault tolerance and reliability limit application scalability. Current techniques to ensure progress across faults, like checkpointrestart, are unsuitable at these scale due to excessive overheads predicted to more than double an applications time to solution. Redundant computation, long used in distributed and mission critical systems, has been suggested as an alternative to checkpoint-restart on its own. In this paper we describe the rMPI library which enables portable and transparent redundant computation for MPI applications. We detail the design of the library as well as two replica consistency protocols, outline the overheads of this library at scale on a number of real-world applications, and finally outline the significant increase in an applications time to solution at extreme scale as well as show the scenarios in which redundant computation makes sense.

Linux clusters have become very popular for scientific com- puting at research institutions world-wide, because they can be easily deployed at a fairly low cost. However, the most pressing issues of today's cluster solutions are availability and serviceability. The conventional Beowulf cluster archi- tecture has a single head node connected to a group of com- pute nodes. This head node

Comparisons of high-performance computers based on their peak floating point performance are common but seldom useful when comparing performance on real workloads. Factors that influence sustained performance extend beyond a system's floating-point units, and real applications exercise machines in complex and diverse ways. Even when it is possible to compare systems based on their performance, other considerations affect which machine is best for a given organization. These include the cost, the facilities requirements (power, floorspace, etc.), the programming model, the existing code base, and so on. This paper describes some of the important measures for evaluating high-performance computers. We present data for many of these metrics based on our experience at Lawrence Livermore National Laboratory (LLNL), and we compare them with published information on the Earth Simulator. We argue that evaluating systems involves far more than comparing benchmarks and acquisition costs. We show that evaluating systems often involves complex choices among a variety of factors that influence the value of a supercomputer to an organization, and that the high-end computing community should view cost/performance comparisons of different architectures with skepticism.

With the advent of Cloud computing, large-scale virtualized compute and data centers are becoming common in the computing industry. These distributed systems leverage commodity server hardware in mass quantity, similar in theory to many of the fastest Supercomputers in existence today. However these systems can consume a cities worth of power just to run idle, and require equally massive cooling systems to keep the servers within normal operating temperatures. This produces CO2 emissions and significantly contributes to the growing environmental issue of Global Warming. Green computing, a new trend for high-end computing, attempts to alleviate this problem by delivering both high performance and reduced power consumption, effectively maximizing total system efficiency.

The last decade has witnessed a rapid proliferation of superscalar cache-based microprocessors to build high-end computing (HEC) platforms, primarily because of their generality, scalability, and cost effectiveness. However, the growing gap between sustained and peak performance for full-scale scientific applications on conventional supercomputers has become a major concern in high per-

The last decade has witnessed a rapid proliferation of superscalar cache-based microprocessors to build high-end computing (HEC) platforms, primarily because of their generality, scalability, and cost effectiveness. However, the growing gap between sustained and peak performance for full-scale scientific applications on conventional supercomputers has become a major concern in high performance computing, requiring significantly larger systems and application scalability than implied by peak performance in order to achieve desired performance. The latest generation of custom-built parallel vector systems have the potential to address this issue for numerical algorithms with sufficient regularity in their computational structure. In this work we explore applications drawn from four areas: magnetic fusion (GTC), plasma physics (LBMHD3D), astrophysics (Cactus), and material science (PARATEC). We compare performance of the vector-based Cray X1, X1E, Earth Simulator, NEC SX-8, with performance of three leading commodity-based superscalar platforms utilizing the IBM Power3, Intel Itanium2, and AMD Opteron processors. Our work makes several significant contributions: a new data-decomposition scheme for GTC that (for the first time) enables a breakthrough of the Teraflop barrier; the introduction of a new three-dimensional Lattice Boltzmann magneto-hydrodynamic implementation used to study the onset evolution of plasma turbulence that achieves over 26Tflop/s on 4800 ES processors; the highest per processor performance (by far) achieved by the full-production version of the Cactus ADM-BSSN; and the largest PARATEC cell size atomistic simulation to date. Overall, results show that the vector architectures attain unprecedented aggregate performance across our application suite, demonstrating the tremendous potential of modern parallel vector systems.

Achieving good performance on high-end computing systems is growing ever more challenging due to enormous scale, increasing architectural complexity, and increasing application complexity. To address these challenges in DOE's SciDAC-2 program , the Performance Engineering Research Institute (PERI) has embarked on an ambitious research plan encompassing performance modeling and prediction, automatic performance optimization and performance engineering of high profile applications. The principal new component is a research activity in automatic tuning software, which is spurred by the strong user preference for automatic tools.

1] The increasing capability of high-end computers allows numerical simulations with horizontal resolutions high enough to resolve cloud systems in a global model. In this paper, initial results from the global Nonhydrostatic ICosahedral Atmospheric Model (NICAM) are highlighted to demonstrate the beginning of a potentially new era for weather and climate predictions with global cloud-systemresolving models. The NICAM simulation with a horizontal resolution of about 7 km successfully reproduced the lifecycles of two real tropical cyclones that formed in Indian Ocean in the austral summer 2006. Initialized with the atmospheric conditions 1-2 weeks before the cyclones genesis, the model captured reasonably not only the timing of the observed cyclone geneses but also their motions and mesoscale structures. The model provides a high temporal/spatial resolution dataset for detailed studies of mesoscale aspects of tropical cyclone genesis. These promising results suggest the predictability of tropical cyclones by high-resolution global cloud-system-resolving models. Citation: Fudeyasu, H., Y. Wang, M. Satoh, T. Nasuno, H. Miura, and W. Yanase (2008), Global cloud-system-resolving model NICAM successfully simulated the lifecycles of two real tropical cyclones, Geophys. Res. Lett., 35, L22808,

We present a de novo hierarchical simulation framework for first-principles based predictive simulations of materials and their validation on high-end parallel supercomputers and geographically distributed clusters. In this framework, highend chemically reactive and non-reactive molecular dynamics (MD) simulations explore a wide solution space to discover microscopic mechanisms that govern macroscopic material properties, into which highly accurate quantum mechanical (QM) simulations are embedded to validate the discovered mechanisms and quantify the uncertainty of the solution. The framework includes an embedded divide-andconquer (EDC) algorithmic framework for the design of linear-scaling simulation algorithms with minimal bandwidth complexity and tight error control. The EDC framework also enables adaptive hierarchical simulation with automated model transitioning assisted by graph-based event tracking. A tunable hierarchical cellular decomposition parallelization framework then maps the O(N) EDC algorithms onto petaflops computers, while achieving performance tunability through a hierarchy of parameterized cell data/ computation structures, as well as its implementation using hybrid grid remote procedure call + message passing + threads programming. High-end computing platforms such as IBM BlueGene/L, SGI Altix 3000 and the NSF TeraGrid provide an excellent test grounds for the framework. On these platforms, we have achieved unprecedented scales of quantum-mechanically accurate and well validated, chemi-cally reactive atomistic simulations-1.06 billion-atom fast reactive force-field MD and 11.8 million-atom (1.04 trillion grid points) quantum-mechanical MD in the framework of the EDC density functional theory on adaptive multigridsin addition to 134 billion-atom non-reactive space-time multiresolution MD, with the parallel efficiency as high as 0.998 on 65,536 dual-processor BlueGene/L nodes. We have also achieved an automated execution of hierarchical QM/MD simulation on a grid consisting of 6 supercomputer centers in the US and Japan (in total of 150,000 processor hours), in which the number of processors change dynamically on demand and resources are allocated and migrated dynamically in response to faults. Furthermore, performance portability has been demonstrated on a wide range of platforms such as BlueGene/L, Altix 3000, and AMD Opteron-based Linux clusters. Downloaded from − − −4 of the bulk density. (c) Energy conservation in EDC-DFT based MD simulation of liquid Rb at 1400 K. The domain and buffer sizes are 16.424 and 8.212 a.u., respectively.

Performance and power are critical design constraints in today's high-end computing systems. Reducing power consumption without impacting system performance is a challenge for the HPC community. We present a runtime system (CPU MISER) and an integrated performance model for performance-directed, power-aware cluster computing. CPU MISER supports system-wide, applicationindependent, fine-grain, dynamic voltage and frequency scaling (DVFS) based power management for a generic power-aware cluster. Experimental results show that CPU MISER can achieve as much as 20% energy savings for the NAS parallel benchmarks. In addition to energy savings, CPU MISER is able to constrain performance loss for most applications within user-specified limits. These constraints are achieved through accurate performance modeling and prediction, coupled with advanced control techniques.

In this paper, we propose and present the design and initial development of the Fault awareness Enabled Computing Environment (FENCE) system for high end computing. FENCE is a comprehensive fault management system in the sense that it consists of both post and runtime analysis, integrates both proactive and reactive mechanisms, and combines both application level and system level fault management. Component-based systems are also developed to support the comprehensive FENCE design. Preliminary implementation results are presented.

Mobile devices, which are progressively surrounded in our everyday life, have created a new paradigm where they interconnect, interact and collaborate with each other. This network can be used for flexible and secure coordinated sharing. On the other hand Grid computing provides dependable, consistent, pervasive, and inexpensive access to high-end computational capabilities. In this paper, efforts are made to map the concepts of Grid on Ad-Hoc networks because both exhibit similar kind of characteristics like Scalability, Dynamism and Heterogeneity. In this context we propose "Mobile Ad-Hoc Services Grid -MASGRID".

K42 is an open-source scalable research operating system well suited to support systems research. The primary goals of K42's design that support such research include flexibility to allow a multitude of policies and implementations to be supported simultaneously, extensibility to allow new policies and implementations to be readily added, and scalability to enable good performance for both small and large applications on both small and large multiprocessor systems. The goals are accomplished via key features including an objectoriented structure that allows specialized resource management implementations and policies on a per-resource, perapplication basis, implementation in user-level servers of much of the system functionality, and a sophisticated set of underlying services that provides a programming model for developing system software in a scalable and modular fashion.

The last decade has witnessed a rapid proliferation of superscalar cache-based microprocessors to build high-end computing (HEC) platforms, primarily because of their generality, scalability, and cost effectiveness. However, the growing gap between sustained and peak performance for full-scale scientific applications on conventional supercomputers has become a major concern in high per-

Linux currently plays an important role in high-end computing systems, but recent work has shown that Linux-related processing costs and variablity in network processing times can limit the scalability of HPC applications. Measuring and understanding these overheads is thus key for future use of Linux in large scale HPC systems. Unfortunately, currently available performance monitoring systems introduce large overheads, performance data is generally not available on-line or from the operating system, and the data collected by such systems is generally coarse-grained. In this paper, we present a low-overhead framework for solving one of these problems: making useful operating system performance data available to the application at runtime. Specifically, we have enhanced Linux Trace Toolkit(LTT) to monitor the performance characteristics of individual system calls and to make per-request performance data available to the application. We demonstrate the ability of this framework to monitor individual network and disk requests, and show that the overhead of our per-request performance monitoring framework is minimal. We also present preliminary measurements of Linux system call overhead on a simple HPC.

A grid consists of high-end computational, storage, and network resources that, while known a priori, are dynamic with respect to activity and availability. Efficient scheduling of requests to use grid resources must adapt to this dynamic environment while meeting administrative policies. In this paper, we describe a framework called SPHINX that can administrate grid policies, and schedule complex and data

The growing gap between sustained and peak performance for scientific applications is a well-known problem in high end computing. The recent development of parallel vector systems offers the potential to bridge this gap for many computational science codes and deliver a substantial increase in computing capabilities. This paper examines the intranode performance of the NEC SX-6 vector processor and the cache-based IBM Power3/4 superscalar architectures across a number of scientific computing areas. First, we present the performance of a microbenchmark suite that examines low-level machine characteristics. Next, we study the behavior of the NAS Parallel Benchmarks. Finally, we evaluate the performance of several scientific computing codes. Results demonstrate that the SX-6 achieves high performance on a large fraction of our applications and often significantly outperforms the cache-based architectures. However, certain applications are not easily amenable to vectorization and would require extensive algorithm and implementation reengineering to utilize the SX-6 effectively. * Employee of Computer Sciences Corporation.

We present a de novo hierarchical simulation framework for first-principles based predictive simulations of materials and their validation on high-end parallel supercomputers and geographically distributed clusters. In this framework, highend chemically reactive and non-reactive molecular dynamics (MD) simulations explore a wide solution space to discover microscopic mechanisms that govern macroscopic material properties, into which highly accurate quantum mechanical (QM) simulations are embedded to validate the discovered mechanisms and quantify the uncertainty of the solution. The framework includes an embedded divide-andconquer (EDC) algorithmic framework for the design of linear-scaling simulation algorithms with minimal bandwidth complexity and tight error control. The EDC framework also enables adaptive hierarchical simulation with automated model transitioning assisted by graph-based event tracking. A tunable hierarchical cellular decomposition parallelization framework then maps the O(N) EDC algorithms onto petaflops computers, while achieving performance tunability through a hierarchy of parameterized cell data/ computation structures, as well as its implementation using hybrid grid remote procedure call + message passing + threads programming. High-end computing platforms such as IBM BlueGene/L, SGI Altix 3000 and the NSF TeraGrid provide an excellent test grounds for the framework. On these platforms, we have achieved unprecedented scales of quantum-mechanically accurate and well validated, chemi-cally reactive atomistic simulations-1.06 billion-atom fast reactive force-field MD and 11.8 million-atom (1.04 trillion grid points) quantum-mechanical MD in the framework of the EDC density functional theory on adaptive multigridsin addition to 134 billion-atom non-reactive space-time multiresolution MD, with the parallel efficiency as high as 0.998 on 65,536 dual-processor BlueGene/L nodes. We have also achieved an automated execution of hierarchical QM/MD simulation on a grid consisting of 6 supercomputer centers in the US and Japan (in total of 150,000 processor hours), in which the number of processors change dynamically on demand and resources are allocated and migrated dynamically in response to faults. Furthermore, performance portability has been demonstrated on a wide range of platforms such as BlueGene/L, Altix 3000, and AMD Opteron-based Linux clusters. Downloaded from − − −4 of the bulk density. (c) Energy conservation in EDC-DFT based MD simulation of liquid Rb at 1400 K. The domain and buffer sizes are 16.424 and 8.212 a.u., respectively.

Failure management is crucial for high performance computing systems, especially when the complexity of applications and underlying infrastructure has grown sharply in recent years. In this paper, we present the design, implementation and experiment of Trouble Dashboard (TD), an adaptive, flexible, and low overhead failure monitoring system. Our goal is to provide a lightweight, scalable failure-monitoring tool for both application scientists and system managers. In TD, a set of APIs is provided for application scientists to control the behavior of their applications with flexibility when failures happen. System managers can use the tool to monitor the status of not only computing nodes and running tasks but also failures when they occur. Experiments show that TD incurs low overhead, and remains accurate and flexible enough to adapt to various applications.

Grid workflows can be seen as special scientific workflows involving high performance and/or high throughput computational tasks. Much work in grid workflows has focused on improving application performance through schedulers that optimize the use of computational resources and bandwidth. As high-end computing resources are becoming more of a commodity that is available to new scientific communities, there is an increasing need to also improve the design and reusability "performance" of scientific workflow systems. To this end, we are developing a framework that supports the design and reuse of grid workflows. Individual workflow components (e.g., for data movement, database querying, job scheduling, remote execution etc.) are abstracted into a set of generic, reusable tasks. Instantiations of these common tasks can be functionally equivalent atomic components (called actors) or composite components (so-called composite actors or subworkflows). In this way, a grid workflow designer does not have to commit to a particular Grid technology when developing a scientific workflow; instead different technologies (e.g. GridFTP, SRB, and scp) can be used interchangeably and in concert. We illustrate the application of our framework using two real-world Grid workflows from different scientific domains, i.e., cheminformatics and bioinformatics, respectively.

In 2003, the High End Computing Revitalization Task Force designated file systems and I/O as an area in need of national focus. The purpose of the High End Computing Interagency Working Group (HECIWG) is to coordinate government spending on File Systems and I/O (FSIO) R&D by all the government agencies that are involved in High End Computing. The HECIWG tasked a smaller advisory group to list, categorize, and prioritize HEC I/O and File Systems R&D needs. In 2005, leaders in FSIO from academia, industry and government agencies collaborated to list and prioritize areas of research in HEC FSIO. This led to a very successful High End Computing University Research Activity (HECURA) call from NSF in 2006 and has prompted a new HECURA call from NSF in 2009. This paper serves as both a review of the 2008 HEC FSIO identified research gaps as well as a preview of this forthcoming HECURA call.

For parallel applications running on high-end computing systems, which processes of an application get launched on which processing cores is typically determined at application launch time without any information about the application characteristics. As high-end computing systems continue to grow in scale, however, this approach is becoming increasingly infeasible for achieving the best performance. For example, for systems such as IBM Blue Gene and Cray XT that rely on flat 3D torus networks, process communication often involves network sharing, even for highly scalable applications. This causes the overall application performance to depend heavily on how processes are mapped on the network. In this paper, we first analyze the impact of different process mappings on application performance on a massive Blue Gene/P system. Then, we match this analysis with application communication patterns that we allow applications to describe prior to being launched. The underlying process management system can use this combined information in conjunction with the hardware characteristics of the system to determine the best mapping for the application. Our experiments study the performance of different communication patterns, including 2D and 3D nearest-neighbor communication and structured Cartesian grid communication. Our studies, that scale up to 131,072 cores of the largest BG/P system in the United States (using 80% of the total system size), demonstrate that different process mappings can show significant difference in overall performance, especially on scale. For example, we show that this difference can be as much as 30% for P3DFFT and up to twofold for HALO. Through our proposed model, however, such differences in performance can be avoided so that the best possible performance is always achieved.

The Information Sciences Institute and Caltech are enabling USJFCOM and the Institute for Defense Analyses to conduct entity-level simulation experiments using hundreds of distributed computer nodes on Linux Clusters as a vehicle for simulating millions of JSAF entities. Included below is the experience with the design and implementation of the code that increased scalability, thereby enabling two orders of magnitude growth and the effective use of DoD high-end computers. A typical JSAF experiment generates several terabytes of logged data, which is queried in near-real-time and for months afterward. The amount of logged data and the desired database query performance mandated the redesign of the original logger system's monolithic database, making it distributed and incorporating several advanced concepts. System procedures and practices were established to reliably execute the globalscale simulations, effectively operate the distributed computers, efficiently process and store terabytes of data, and provide straightforward access to the data by analysts.

In order to take full advantage of high-end computing platforms, scientific applications often require modifications to source codes, and to their build systems that generate executable files. The ever-increasing emphasis on productivity in HPC makes the efficiency of the build process an extremely important issue. Our work is motivated by the need for a systematic approach to the HPC lifecycle that encompasses build and porting tasks. In this paper we briefly present the design of the Harness Workbench Toolkit (HWT) that addresses the porting issue through a user-assisted conversion process. We briefly describe our adaptation capability model that includes routine conversions (string substitutions) as well as more advanced transformations such as 32-64 bit changes. Next, we show and discuss results of an experiment with a production source code (the CPMD application) that examines the effort of adapting the baseline code (the Linux distribution) to specific high-end machines (IBM SP4, Cray X1E, Cray XT3/4) in terms of the number of necessary conversions. Based on the conversion capability model, we have implemented conversion assistant modules that were used in the experiment. The experimental results are promising and demonstrate that our approach takes a step towards improving the overall productivity of scientific applications on high-end machines.