Loading…

Sign up or log in to see what your friends are attending and create your own schedule!


View analytic
Filter: Camelot 3rd Floor
 

10:00am

Tech: Gordon: Design, Performance, and Experiences
    Tuesday July 17, 2012 10:00am - 10:30am @ Camelot 3rd Floor

    Tech: Gordon: Design, Performance, and Experiences Deploying and Supporting a Data-Intensive Supercomputer

    Abstract: The Gordon data intensive supercomputer entered service in early 2012 as an allocable computing system in the NSF Extreme Science and Engineering Discovery Environment (XSEDE) program. Gordon has several innovative features that make it ideal for data intensive computing including: 1,024, dual socket, 16-core, 64GB compute nodes based on Intel’s Sandy Bridge processor; 64 I/O nodes with an aggregate of 300 TB of high performance flash (SSD); large, virtual SMP “supernodes” of up to 2 TB DRAM; a dual-rail, QDR InfiniBand, 3D torus network based on commodity hardware and open source software; and a 100 GB/s Lustre based parallel file system, with over 4 PB of disk space. In this paper we present the motivation, design, and performance of Gordon. We provide: low level micro-benchmark results to demonstrate processor, memory, I/O, and network performance; standard HPC benchmarks; and performance on data intensive applications to demonstrate Gordon’s performance on typical workloads. We highlight the inherent risks in, and offer mitigation strategies for, deploying a data intensive supercomputer like Gordon which embodies significant innovative technologies. Finally we present our experiences thus far in supporting users and managing a system like Gordon.

     



    Speakers

    Type Technology Track
    Session Titles XSEDE Service Provider Systems


10:30am

Tech: A Tale of Two Systems
    Tuesday July 17, 2012 10:30am - 11:00am @ Camelot 3rd Floor

    Tech: A Tale of Two Systems: Flexibility of Usage of Kraken and Nautilus at the National Institute for Computational Sciences

    Abstract: The National Institute for Computational Sciences (NICS) currently operates two computational resources for the eXtreme Science and Engineering Discovery Environment (XSEDE), Kraken, a 112896-core Cray XT5 for general purpose computation, and Nautilus, a 1024-core SGI Altix UV 1000 for data analysis and visualization. We analyze a year worth of accounting logs for Kraken and Nautilus to understand how users take advantage of these two systems and how analysis jobs differ from general HPC computation We find that researchers take advantage of the flexibility offered by these sytems, running a wide variety of jobs at many scales and using the full range of core counts and available memory for their jobs. The jobs on Nautilus tend to use less walltime and more memory per core than the jobs run on Kraken. Additionally, researchers are more likely to run interactive jobs on Nautilus than on Kraken. Small jobs experience a good quality of service on both systems. This information can be used for the management and allocation of time on existing HPC and analysis systems as well as for planning for deploying future HPC and analysis systems.

     



    Speakers

    Type Technology Track
    Session Titles XSEDE Service Provider Systems


11:00am

Tech: Analyzing .. Trestles
    Tuesday July 17, 2012 11:00am - 11:30am @ Camelot 3rd Floor

    Tech: Analyzing Throughput and Utilization on Trestles

    Abstract: The Trestles system is targeted to modest-scale and gateway users, and is operated to enhance users’ productivity by maintaining good turnaround time as well as other user-friendly features such as long run times and user reservations. However the goal of maintaining good throughput competes with the goal of high system utilization. This paper analyzes one year of Trestles operations to characterize the empirical relationship between utilization and throughput, with the objectives of understanding their trade-off, and informing allocations and scheduling policies to optimize this trade-off. There is considerable scatter in the correlation between utilization and throughput, as measured by expansion factor. There are periods of good throughput at both low and high utilizations, while there are other periods when throughput degrades significantly not only at high utilization but even at low utilization. However, throughput consistently degrades above ~90% utilization. User behavior clearly impacts the expansion factor metrics; the great majority of jobs with extreme expansion factors are associated with a very small fraction of users who either flood the queue with many jobs or request run times far in excess of actual run times. While the former is a user workflow choice, the latter clearly demonstrates the benefit for users to request run times that are well-matched to actual run times. Utilization and throughput metrics derived from XDMoD are compared for Trestles with two other XSEDE systems, Ranger and Kraken, with different sizes and allocation/scheduling policies. Both Ranger and Kraken have generally higher utilization and, not surprisingly, higher expansion factors than Trestles over the analysis period. As a result of this analysis, we intend to increase the target allocation fraction from the current 70% to ~75-80%, and strongly advise users to reasonably match requested run times to actual run times.

     



    Speakers

    Type Technology Track
    Session Titles XSEDE Service Provider Systems


11:30am

Tech: Invited Talk: UNICORE 6
    Tuesday July 17, 2012 11:30am - 12:00pm @ Camelot 3rd Floor

    Tech: Invited Talk: UNICORE 6 in XSEDE

    Abstract: UNICORE (Uniform Interface to Computing Resources) offers a ready-to-run Grid system including client and server software. UNICORE makes distributed computing and data resources available in a seamless and secure way in intranets and the internet. UNICORE 6 is deployed at PRACE sites all over Europe, as well as in the D-Grid (Deutsches Grid). UNICORE 6 implementations have been demonstrated to work during the XSEDE proposal process and currently XSEDE has been working towards deployment of UNICORE through the XSEDE Software Development and Integration process. An overview of UNICORE for XSEDE and the current status of the development activities and deployment will be discussed.

     



    Speakers

    Type Technology Track
    Session Titles XSEDE Service Provider Systems


2:45pm

Science: Three-dimensional Simulations
    Tuesday July 17, 2012 2:45pm - 3:15pm @ Camelot 3rd Floor

    Science: Three-dimensional Simulations of Geometrically Complex Subduction with Large Viscosity Variations

    Abstract: The incorporation of geologic realism into numerical models of subduction is becoming increasingly necessary as observational and experimental constraints indicate plate boundaries are inherently three-dimensional (3D) in nature and contain large viscosity variations. However, large viscosity variations occurring over short distances pose a challenge for computational codes, and models with complex 3D geometries require substantially greater numbers of elements, increasing the computational demands. We modified a community mantle convection code, CitcomCU, to model realistic subduction zones that use an arbitrarily shaped 3D plate boundary interface and incorporate the effects of a strain-rate dependent viscosity based on an experimentally derived flow law for olivine aggregates. Tests of this implementation on 3D models with a simple subduction zone geometry indicate that limiting the overall viscosity range in the model, as well as limiting the viscosity jump across an element, improves model runtime and convergence behavior, consistent with what has been shown previously. In addition, the choice of interpolation method and averaging scheme used to transfer the viscosity structure to the different levels in the multigrid solver can significantly improve model performance. These optimizations can improve model runtime by over 20%. 3D models of a subduction zone with a complex plate boundary geometry were then constructed, containing over 100 million finite element nodes with a local resolution of up to 2.35 km, and run on the TeraGrid. These complex 3D models representative of the Alaska subduction zone-transform plate boundary contain viscosity variations of up to seven orders of magnitude. The optimizations in solver parameters determined from the simple 3D models of subduction applied to the much larger and more complex models of an actual subduction zone improved model convergence behavior and reduced runtimes by on the order of 25%. One scientific result from 3D models of Alaska is that a laterally variable mantle viscosity emerges in the mantle as a consequence of variations in the flow field, with localized velocities of greater than 80 cm/yr occurring close to the subduction zone where the negative buoyancy of the slab drives the flow. These results are a significant departure from the paradigm of two-dimensional (2D) models of subduction where the slab velocity is often fixed to surface plate motion. While the solver parameter optimization can improve model performance, the results also demonstrate the need for new solvers to keep pace with the demands for increasingly complex numerical simulations in mantle convection.

     



    Speakers

    Type Science Track
    Session Titles Fluid Dynamics


3:15pm

Science: Multiscale Modeling
    Tuesday July 17, 2012 3:15pm - 3:45pm @ Camelot 3rd Floor

    Science: Multiscale Modeling of High Explosives for Transportation Accidents

    Abstract: The development of a reaction model to simulate the accidental detonation of a large array of seismic boosters in a semi-truck subject to fire is considered. To test this model large scale simulations of explosions and detonations were performed by leveraging the massively parallel capabilities of the Uintah Computational Framework and the XSEDE computational resources. Computed stress profiles in bulk-scale explosive materials were validated using compaction simulations of hundred micron scale particles and found to compare favorably with experimental data. A validation study of reaction models for deflagration and detonation showed that computational grid cell sizes up to 10 mm could be used without loss of fidelity. The Uintah Computational Framework shows linear strong scaling up to 180K cores which combined with coarse resolution and validated models will now enable simulations of semi-truck scale transportation accidents for the first time.

     



    Speakers

    Type Science Track
    Session Titles Fluid Dynamics


3:45pm

Science: Multiscale simulations
    Tuesday July 17, 2012 3:45pm - 4:15pm @ Camelot 3rd Floor

    Science: Multiscale simulations of Langmuir cells and submesoscale eddies using XSEDE resources

    Abstract: A proper treatment of upper ocean mixing is an essential part of accurate climate modeling. This problem is difficult because the upper ocean is home to many competing processes. Vertical turbulent mixing acts to unstratify the water column, while lateral submesoscale eddies attempt to stratify the column. Langmuir turbulence, which often dominates the vertical mixing, is driven by an interaction of the wind stress and surface wave (Stokes) drift, while the submesoscale eddies are driven by lateral density and velocity changes. Taken together, these processes span a large range of spatial and temporal scales. They have been studied separately via theory and modeling. It has been demonstrated that the way these scales are represented in climate models has a nontrivial impact on the global climate system. The largest impact is on upper ocean processes, which filter air-sea interactions. This interaction is especially interesting, because it is the interface between nonhydrostatic and hydrostatic, quasigeostrophic and ageostrophic, and small-scale and large-scale ocean dynamics. Previous studies have resulted in parameterizations for Langmuir turbulence and submesoscale fluxes, but these parameterizations assume that there is no interaction between these important processes. In this work we have utilized a large XSEDE allocation (9 million SUs) to perform multiscale simulations that encompass the Langmuir Scale (O(10-100m)) and the submesoscale eddies (O(1-10km)). One simulation includes a Stokes drift, and hence Langmuir turbulence, while the other does not. 
    To adequately represent such disparate spatial scales is a challenge in numerous regards. Numerical prediction algorithms must balance efficiency, scalability, and accuracy. These simulations also present a large challenge for data storage and transfer. However, the results of these simulations will influence climate modeling for decades.



    Speakers

    Type Science Track
    Session Titles Fluid Dynamics


4:45pm

BOF: Stampede Alert!
    Tuesday July 17, 2012 4:45pm - 5:45pm @ Camelot 3rd Floor

    BOF: Stampede Alert! Come Contribute or Get Outta the Way

    Abstract: Stampede, the next big HPC system in the XSEDE program, will go into production in January 2013. Stampede will be tremendously powerful computational platform, leveraging Dell nodes containing Intel Sandy Bridge processors and forthcoming MIC coprocessors to provide 10PF peak performance. Stampede will also have tremendous memory, disk, and visualization capabilities, a set of large shared memory nodes, software that enables high throughput computing, excellent interconnect latency and bandwidth, a rich set of software and services, and outreach efforts including campus bridging efforts to help other sites deploy MIC-based clusters. Most importantly, the Stampede project is designed to support hundreds of diverse science applications and requirements spanning domains and usage models. We invite you to come learn about the system and project, and to provide your suggestions for how we can deliver the most productive system and services for the open science community.

     



    Speakers

    Type BOF


 
 

10:00am

Tech: Radiation Modeling
    Wednesday July 18, 2012 10:00am - 10:30am @ Camelot 3rd Floor

    Tech: Radiation Modeling Using the Uintah Heterogeneous CPU/GPU Runtime System

    Abstract: The Uintah Computational Framework was developed to provide an environment for solving fluid-structure interaction problems on structured adaptive grids on large-scale, long-running, data-intensive problems. Uintah uses a combination of fluid-flow solvers and particle-based methods for solids, together with a novel asynchronous task-based approach with fully automated load balancing. Uintah demonstrates excellent weak and strong scalability at full machine capacity on XSEDE resources such as Ranger and Kraken, and through the use of a hybrid memory approach based on a combination of MPI and Pthreads, Uintah now runs on up to 262k cores on the DOE Jaguar system. In order to extend Uintah to heterogeneous systems, with ever-increasing CPU core counts and additional on-node GPUs, a new dynamic CPU-GPU task scheduler is designed and evaluated in this study. This new scheduler enables Uintah to fully exploit these architectures with support for asynchronous, out-of-order scheduling of both CPU and GPU computational tasks. A new runtime system has also been implemented with an added multi-stage queuing architecture for efficient scheduling of CPU and GPU tasks. This new runtime system automatically handles the details of asynchronous memory copies to and from the GPU and introduces a novel method of pre-fetching and preparing GPU memory prior to GPU task execution. In this study this new design is examined in the context of a developing, hierarchical GPU-based ray tracing radiation transport model that provides Uintah with additional capabilities for heat transfer and electromagnetic wave propagation. The capabilities of this new scheduler design are tested by running at large scale on the modern heterogeneous systems, Keeneland and TitanDev, with up to 360 and 960 GPUs respectively. On these systems, we demonstrate significant speedups per GPU against a standard CPU core for our radiation problem.

     



    Speakers

    Type Technology Track
    Session Titles GPUs and Software


10:30am

Tech: An Analysis of GPU
    Wednesday July 18, 2012 10:30am - 11:00am @ Camelot 3rd Floor

    Tech: An Analysis of GPU Utilization Trends on the Keeneland Initial Delivery System

    Abstract: In late 2010, The Georgia Institute of Technology along with its partners – the Oak Ridge National Lab, the University of Tennessee-Knoxville, and the National Institute for Computational Sciences, deployed the Keeneland Initial Delivery System (KIDS) - a 201 Teraflop, 120-node HP SL390 system with 240 Intel Xeon CPUs and 360 NVIDIA Fermi graphics processors as a part of the Keeneland Project. The Keeneland Project is a five-year Track 2D cooperative agreement awarded by the National Science Foundation (NSF) in 2009 for the deployment of an innovative high performance computing system in order to bring emerging architectures to the open science community and KIDS is being used to develop programming tools and libraries in order to ensure that the project can productively accelerate important scientific and engineering applications. Until late 2011, there was no formal mechanism in place for quantifying the efficiency of GPU usage on the Keeneland system because most applications did not have the appropriate administrative tools and vendor support. GPU administration has largely been an afterthought as vendors in this space are focused on gaming and video applications. There is a compelling need to monitor GPU utilization on Keeneland for the purposes of proper system administration and future planning for Keeneland Final System, which is expected to be in production in July 2012. With the release of CUDA 4.1, NVIDIA added enhanced functionality to the nvidia-system management interface (nvidiasmi) tool, which is a management and monitoring command line utility that leverages the NVIDIA Management Library (NVML). NVML is a C-based API for monitoring and managing various states of the NVIDIA GPU devices. It provides a direct access to the queries and commands exposed via nvidia-smi. Using nvidiasmi, a monitoring tool was built for KIDS, to monitor utilization and memory usage on the GPUs. In this paper, we discuss the development of the GPU Utilization tool in depth, and its implementation details on KIDS. We also provide an analysis of the utilization statistics generated by this tool. For example, we identify utilization trends across jobs submitted on KIDS – such as overall GPU utilization as compared to CPU utilization (figure 1), and we investigate how GPU utilization changes for different job sizes with changes in GPU hours requested. We also examine GPU utilization from the perspective of software – which packages are most frequently used, and how do they compare with respect to GPU utilization (figure 2) and memory usage. Collection and analysis of this data is essential for facilitating heterogeneous computing on the Keeneland Initial Delivery System. Future direction for the usage of these statistics is to provide insights on overall usage of the system, determine appropriate ratios for jobs (CPU to GPU, GPU to host memory), assist in scheduling policy management, and determine software utilization. These statistics become even more relevant as the center prepares for the deployment of the Keeneland Final System. As heterogeneous computing appears to be more and more common, and is quickly becoming the standard, this information will help greatly in delivering consistent high uptime and assist software developers in writing more efficient code for the majority of the codebases aimed at heterogeneous systems.



    Speakers

    Type Technology Track
    Session Titles GPUs and Software


11:00am

Tech: A Distributed Memory
    Wednesday July 18, 2012 11:00am - 11:30am @ Camelot 3rd Floor

    Tech: A Distributed Memory Out-of-Core Method and Its Application to Quantum Chemistry Applications

    Abstract: Out-of-core methods, which repeatedly offload data to disk in order toovercome local on-node memory constraints are encountered in a rangeof scientific computing disciplines, including quantumchemistry. Unfortunately, these methods do not often map nicely ontoglobal parallel file systems employed on modern HPC clusters and canoverwhelm even the most capable of file systems causing unacceptablylow application performance (while also degrading I/O performance forall system users). To address this bottleneck and explore moreefficient use of HPC clusters for a quantum chemistry application,CFOUR, a new MPI-based utility has been developed to supportout-of-core methods on distributed memory systems. This MPI Ocore utility leverages the high-speed interconnect available onHPC clusters to offload and retrieve out-of-core records to one ormore remote memory storage pools, avoiding excessive I/O transactionson local or global file systems. In this paper, we present an overviewof the Ocore implementation, it's direct application within a largequantum chemistry application, and micro-benchmark and applicationperformance results from an HPC cluster interconnected withquad-data-rate InfiniBand.



    Speakers

    Type Technology Track
    Session Titles GPUs and Software


11:30am

Tech: Achieve Better Performance
    Wednesday July 18, 2012 11:30am - 12:00pm @ Camelot 3rd Floor

    Tech: Achieve Better Performance with PEAK on XSEDE Resources

    Abstract: As the leading distributed cyberinfrastructure for open scientific research in the United States, XSEDE supports several supercomputers across the country, as well as computational tools that are critical to the success of those researchers. In most cases, users are looking for a systematic way of selecting and configuring the available systems software and libraries for their applications so as to obtain optimal application performance. However, few scientific application developers have the time for an exhaustive search of all the possible configurations to determine the best one, and performing such a search empirically can consume a significant proportion of their allocation hours. We present here a framework, called the Performance Environment Autoconfiguration frameworK (PEAK), to help developers and users of scientific applications to select the optimal configuration for their application on a given platform and to update that configuration when changes in the underlying hardware and systems software occur. The choices to be made include the compiler with its settings of compiling options, the numerical libraries and settings of library parameters, and settings of other environment variables to take advantage of the NUMA systems. The framework has helped us choose the optimal configuration to get a significant speedup for some scientific applications executed on XSEDE platforms such as Kraken, Ranger, Nautilus and Blacklight.

     



    Speakers

    Type Technology Track
    Session Titles GPUs and Software


1:15pm

Tech: Invited Talk: Improving XSEDE
    Wednesday July 18, 2012 1:15pm - 1:45pm @ Camelot 3rd Floor

    Tech: Invited Talk: Improving XSEDE Software Quality using Software Engineering Best Practice

    Abstract: XSEDE is introducing a range of system and software engineering practices to achieve systematic and continuous improvement in the quality of its integrated and supported software. This paper will describe XSEDE’s software engineering practices and what we hope to obtain from them. We discuss the technical and cultural challenges of establishing community-defined practices, and the techniques we have been using to address these challenges. We will introduce the initial engineering practices implemented in project year 1, outline additional engineering practice improvements planned during project year 2, and suggest how these engineering practices could be leveraged by the broader XSEDE community.

     



    Speakers

    Type Technology Track
    Session Titles Software and Middleware


1:45pm

Tech: A Framework for Federated
    Wednesday July 18, 2012 1:45pm - 2:15pm @ Camelot 3rd Floor

    Tech: A Framework for Federated Two-Factor Authentication Enabling Cost-Effective Secure Access to Distributed Cyberinfrastructure

    Abstract: As cyber attacks become increasingly sophisticated, the security measures used to mitigate the risks must also increase in sophistication.One time password (OTP) systems provide strong authentication because security credentials are not reusable, thus thwarting credential replay attacks. The credential changes regularly, making brute-forceattacks significantly more difficult. In high performance computing,end users may require access to resources housed at several differentservice provider locations. The ability to share a strong token betweenmultiple computing resources reduces cost and complexity. The National Science Foundation (NSF) Extreme Science and EngineeringDiscovery Environment (XSEDE) provides access to digital resources,including supercomputers, data resources, and software tools. XSEDE willoffer centralized strong authentication for services amongst serviceproviders that leverage their own user databases and security profiles.This work implements a scalable framework built on standards to providefederated secure access to distributed cyberinfrastructure.



    Speakers

    Type Technology Track
    Session Titles Software and Middleware


2:15pm

Tech: Running Many Molecular
    Wednesday July 18, 2012 2:15pm - 2:45pm @ Camelot 3rd Floor

    Tech: Running Many Molecular Dynamics Simulations on Many Supercomputers

    Abstract: The challenges facing biomolecular simulations are manyfold. In addition to long time simulations of a single large system, an important challenge is the ability to run a large number of identical copies (ensembles) of the same system. Ensemble-based simulations are important for effective sampling and due to the low-level of coupling between them, ensemble-based simulations are good candidates to utilize distributed cyberinfrastructure. The problem for the practitioner is thus effectively marshaling thousands if not millions of high-performance simulations on distributed cyberinfrastructure. Here we assess the ability of an interoperable and extensible pilot- job tool (BigJob), to support high-throughput simulations of high- performance molecular dynamics simulations across distributed supercomputing infrastructure. Using a nucleosome positioning problem as an exemplar, we demonstrate how we have addressed this challenge on the TeraGrid/XSEDE. Specifically, we compute 336 independent trajectories of 20 ns each. Each trajectory is further divided into twenty 1 ns long simulation tasks. A single task requires ≈ 42 MB of input, 9 hours of compute time on 32 cores, and generates 3.8 GB of data. In total we have 6,720 tasks (6.7 μs ) and approximately 25 TB to manage. There is natural task-level concurrency, as these 6,720 can be executed with 336-way task concurrency. Using NAMD 2.7, this project requires approximately 2 million hours of CPU time and could be completed in just over 1 month on a dedicated supercomputer containing 3,000 cores. In practice even such a modest supercomputer is a shared resource and our experience suggests that a simple scheme to automatically batch queue the tasks, might require several years to complete the project. In order to reduce the total time-to-completion, we need to scale-up, out and across various resources. Our approach is to aggregate many ensemble members into pilot-jobs, distribute pilot-jobs over multiple compute resources concurrently, and dynamically assign tasks across the available resources.

     



    Speakers

    Type Technology Track
    Session Titles Software and Middleware


2:45pm

Tech: A Systematic Process
    Wednesday July 18, 2012 2:45pm - 3:15pm @ Camelot 3rd Floor

    Tech: A Systematic Process for Efficient Execution on Intel's Heterogeneous Computation Nodes

    Abstract: Heterogeneous architectures (mainstream CPUs with accelerators / co-processors) are expected to become more prevalent in high performance computing clusters. This paper deals specifically with attaining efficient execution on nodes which combine Intel's multicore Sandy Bridge chips with MIC manycore chips. The architecture and software stack for Intel's heterogeneous computation nodes attempt to make migration from the now common multicore chips to the manycore chips straightforward. However, specific execution characteristics are favored by these manycore chips such as making use of the wider vector instructions, minimal inter-thread conflicts, etc. Additionally manycore chips have lower clock speed and no unified last-level cache. As a result, and as we demonstrate in this paper, it will commonly be the case that not all parts of an application will execute more efficiently on the manycore chip than on the multicore chip. This paper presents a process, based on measurements of execution on Westmere-based multicore chips, which can accurately predict which code segments will execute efficiently on the manycore chips and illustrates and evaluates its application to three substantial full programs -- HOMME, MOIL and MILC. The effectiveness of the process is validated by verifying scalability of the specific functions and loops that were recommended for MIC execution on a Knights Ferry computation node.

     



    Speakers

    Type Technology Track
    Session Titles Software and Middleware
    Tags Software and Middleware


3:45pm

Panel: SaSS for Science: The Path to Reality for “Research in the Cloud"
    Wednesday July 18, 2012 3:45pm - 5:15pm @ Camelot 3rd Floor

    Abstract: With the world moving to web-based tools for everything from photo sharing to research publication, it’s no wonder scientists are now seeking online technologies to support their research. But the requirements of large-scale computational research are both unique and daunting: massive data, complex software, limited budgets, and demand for increased collaboration. While “the cloud” promises to alleviate some of these pressures, concerns about feasibility still exist for scientists and the resource providers that support them. 

    This panel will explore the capacity of Software as a Service (SaaS) to transform computational research so the challenges above can be leveraged to advance, not hinder, innovation and discovery. Leaders from each constituency of a scientific research environment (investigator, campus champion, supercomputing facility, SaaS provider) will debate the feasibility of SaaS-based research, examining the delta between current and desired state from a technology and adoptability perspective. We will explore the delta between where we are – and where we need to be – for scientists to reliably and securely perform research in the cloud.

    Panel Moderator: 
    - Ian Foster, Computation Institute, University of Chicago and Argonne

    Panel participants:
    - Nancy Cox, University of Chicago
    - Brock Palen, University of Michigan
    - J. Ray Scott, Pittsburgh Supercomputing Center
    - Steve Tuecke, University of Chicago

     



    Speakers

    Type Panel Session


5:30pm

BOF: Gordon Data Intensive HPC System
    Wednesday July 18, 2012 5:30pm - 6:30pm @ Camelot 3rd Floor

    BOF: Gordon Data Intensive HPC System

    Abstract: This BOF will give current and prospective users the background on Gordon’s architecture, examples of application results obtained on Gordon, and insights in the types of problems that may be well-suited for Gordon. A brief presentation on these topics will help orient the audience, but the majority of the session will be dedicated to a discussion with attendees about their applications and how they might take advantage of Gordon’s unique architectural features. The goal of the BOF is to begin building a community of data intensive application users who can share best practices, and begin to learn from one another. Users from non-traditional HPC domains such as the humanities, social science, and economics are encouraged to attend. This will also be a good opportunity for Campus Champions to learn about Gordon so they are well-prepared to engage their local communities.

     



    Speakers

    Type BOF


 
 

8:45am

Tech: Evaluation of Parallel

9:15am

Tech: Using Kerberized Lustre

9:45am

Tech: The Data Supercell
    Thursday July 19, 2012 9:45am - 10:15am @ Camelot 3rd Floor

    Tech: The Data Supercell

    Abstract: In April of 2012, the Pittsburgh Supercomputing Center unveiled a unique mass storage platform named 'The Data Supercell'. The Data Supercell (DSC) is based on the SLASH2 filesystem, also developed at PSC, and incorporates multiple classes of systems into its environment for the purposes of aiding scientific users and storage administrators. 

    The Data Supercell aims to play a major role in the XSEDE storage ecosystem. Besides serving the vanilla role of PSC's mass storage system, DSC features the new novelties of the SLASH2 filesystem. Outfitted with these new features DSC seeks to provide a new class of integrated storage services to serve users of large scientific data and XSEDE resource providers. 

    This submission will cover all or most aspects of the DSC. These will include:  

    * Software and Hardware Architecture 

    Here will be explained the types of storage systems which compose the DSC with emphasis on the heterogeneous nature of the assembly. Particularly, the SLASH2 I/O service is very portable across system class and operating system. We will detail how this feature was instrumental in constructing the DSC by enabling the inclusion of a legacy tape system with dense storage bricks running ZFS. 

    * Performance Analysis of DSC 

    The community will be interested in the performance of any new distributed / parallel filesystem. Since DSC is the flagship SLASH2 deployment at this time, we shell disseminate I/O performance measurements for data and metadata operations through this submission.  

    * Novel features such as File Replication and Poly-residencies 

    At this time, DSC is primarily used as PSC's mass storage system however the system has interesting capabilities which extend beyond a the features of a traditional archiver. Of these, is the ability to move data in parallel between a scratch filesystem (ie Lustre or GPFS) and the highly dense storage nodes. Further, such features can be enacted by normal users of the system allowing them to transfer data between mass storage and (any) parallel filesystem with exceptional performance. 

     

    SLASH2's file replication capabilities allow for users and administrators to determine the layout, residency, and number of replicas on a per-block basis or for a whole file. Our paper will illuminate such capabilities as used on the DSC. 

    * Upcoming Integrated scientific cloud data services 

    Here we shall describe the how existing and upcoming SLASH2 features will be used to aid XSEDE's large data users. This will focus on users and / or research groups with considerable on-campus storage and compute resources which frequently operate on XSEDE resources. The section will describe in detail the vision of incorporating data replication, user-specific eventually consistent metadata volumes, data multi-residency, and system managed parallel replication for creating tightly integrated storage environments between large scale campus and XSEDE RP resources.

     



    Speakers



    Professor of Physics and Director, Pittsburgh Supercomputing...




    Type Technology Track
    Session Titles File Systems
    Tags File Systems


 

Get Adobe Flash player