archive-edu.com » EDU » U » UCHICAGO.EDU

Total: 7

Choose link from "Titles, links and description words view":

Or switch to "Titles and links view".

  • provided by our successful Beagle 2 grant application The upgrade would consist of Addition of 2 24 PB of raw disk space in two new DDN cabinets for a total of 2 84 PB of raw storage 2 1 PB total usable All compute blades upgraded from 6 core Magny Cours processors to 8 core Abu Dhabi processors increasing core count per node from 24 to 32 All compute blades upgraded from 32GB to 64GB per node Login nodes network cards upgraded from 1Gbps NICs to 10Gbps NICs Adding 4 compute nodes with nVidia GPU processors Work on those hardware upgrades are scheduled to begin on November 11 At that time Beagle will be offline for three weeks while that work is completed Please note that Lustre data will be kept intact as part of the upgrade While Beagle is offline there is no way to access any data on Lustre At this time we expect the work to happen accordingly although these exact dates may still change in the future Week of Nov 10th Beagle is down for pre upgrade prep Week of Nov 17th Beagle hardware upgrade is started Week of Nov 24th Beagle hardware upgrade is completed Week of Dec 1st Beagle acceptance test is completed users are given access to Beagle2 Director The acquisition of the Beagle supercomputer was made possible by a grant from the National Institutes of Health NIH National Center for Research Resources NCRR Ian Foster director of the Computation Institute at the University of Chicago and Argonne National Laboratory is the PI for this project Ian Foster with UChicago s team of technical and domain specialists identified the need for a powerful computational environment that would serve the growing resource intensive requirements of the biomedical research community Beagle s skin was created by the Computation Institute s Mark Hereld and Greg Cross Beagle 2011 is built on three components water and sky are divided by a wave Moving to the right the wave takes on the pitch of the double helix of DNA The images of water and sky are generated by a stochastic context free grammar using a computer This application of stochastic image generation gives Beagle 2011 a fractal aspect that combines visual elements inspired by biology and mathematics disciplines at the heart of the research that Beagle will carry forward System Specifications About 200nd fastest machine Nov 2011 Cray XE6 system 150 Teraflops 600 TB disk 450 TB formatted Extreme Scalability Mode ESM which supports large scalable custom applications Cluster Compatibility Mode CCM which allows standard programs designed for smaller machines or clusters to run without modifications The nodes are connected in a 3D torus topology via the Cray Gemini interconnect A high speed inter processor connection network to support tightly coupled computational simulation and data intensive analysis applications that involve frequent inter process communication At least 32 Gigabyte memory per compute node for applications that create large in memory data structures or that will run many tasks on

    Original URL path: http://beagle.ci.uchicago.edu/ (2014-11-13)
    Open archived version from archive

  • » Events
    from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series High Throughput Computing March 26th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Multicore Madeness March 12th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Applications and Types of Parallelism March 5th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Distributed Multiprocessing February 26th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Shared Memory Multithreading February 19th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Stupid Compiler Tricks February 12th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series Instruction Level Parallelism February 5th from 2 pm to 3 30 pm CT room 236 Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series The Tyranny of the Storage Hierarchy January 29th from 2 pm to 4 pm CT room 236 Searle building SWIFT Intro February 20th 2013 from 1 pm to 3 pm room 240A Searle building Supercomputing in Plain English Spring 2013 A High Performance Computing Workshop Series starts on January 22nd from 2 pm to 4 pm CT 4 00pm Atlantic 3 00pm Eastern 1 00pm Mountain 12 00noon Pacific For FREE registration Send e mail to hneeman ou edu Live via videoconferencing scroll down Overview What the Heck Is Supercomputing So far the SiPE workshops have reached roughly 1000 people at 166 institutions agencies companies and organizations in 42 US states and territories and 5 other countries 125 academic institutions 18 government agencies 16 private firms 7 not for profit organizations SiPE is targeted at an audience of not only computer scientists but especially scientists and engineers including a mixture of undergraduates graduate students faculty and staff These workshops focus on fundamental issues of HPC as they relate to Computational and Data enabled Science Engineering including the storage hierarchy instruction level parallelism high performance compilers shared memory parallelism e g OpenMP distributed parallelism e g MPI HPC application types and parallel paradigms multicore optimization high throughput computing GPGPU computing scientific and I O libraries scientific visualization The key philosophy of the SiPE workshops is that an HPC based code should be maintainable extensible and most especially portable across platforms and should be sufficiently flexible that it can adapt to and adopt emerging HPC paradigms Introduction to High Performance Computing for Biology students January 9th 2013 from 2 30 pm to 4 pm room

    Original URL path: http://beagle.ci.uchicago.edu/trainings-and-events/ (2014-11-13)
    Open archived version from archive

  • » Science on Beagle
    based methods for the optimization of breast tomosynthesis Example of feature extraction followed by dimensionality reduction before the training of a classifier for the identification of breast cancer lesions Credit Giger top The collaborating groups of Karl Freed and Tobin Sosnick University of Chicago use Beagle to develop and validate new protein structure and folding pathway prediction algorithms that start from the amino acid sequence without the use of templates Their approach employs novel compact molecular representations and innovative moves of the protein backbone to achieve accurate prediction with far less computation than previous methods One of their methods called loop modeling which rebuilds local protein structures was highly ranked in the international CASP9 competition Critical Assessment of Protein Structure Prediction The group is using Beagle to evaluate loop modeling algorithm on new proteins and parameter spaces to develop a new tertiary fixing algorithm and to compare these algorithm to other approaches Computational prediction is important for proteins whose structure has not yet been or cannot be determined by experimental methods These largely unsolved problems are crucial for basic biological research and for applications in national priority areas such as pathogens biofuels remediation ecosystems metagenomics crops and nutrition top Tobin Sosnick s group developed and evaluated a new more accurate and efficient predictive methods for large scale explorations of RNA protein interactions yielding new insights into cellular networks Their modFTDock application M Parisien et al can identify protein molecules having a high structural affinity towards a specific RNA molecule Non coding RNAs often function in cells through specific interactions with their protein partners Experiments alone cannot provide a complete picture of this interaction No existing computing method was previously able to provide such genome wide predictions of docked RNA protein complexes Using Beagle the group was able to process 1 500 proteins using the Swift parallel scripting language This work was presented in a poster at the 2011 annual RNA Society meeting Parisien M Sosnick T R Pan T Systematic prediction and validation of RNA protein interactome RNA Society Kyoto June 2011 http www ci uchicago edu wilde RNA2011 poster pdf A manuscript is in progress top The Jinbo Xu s group University of Chicago is developing continued enhancements to a template based method of protein structure modeling that he has embodied in the Raptor family of structure prediction applications These methods give excellent results for proteins where sufficient homologies can be found among proteins of known structure The methods have scored excellently among template based approaches in the international CASP assessment process Results are described in two submitted papers that acknowledge Beagle 1 Jianzhu Ma Jian Peng Sheng Wang and Jinbo Xu A Conditional Neural Fields model for protein threading submitted to RECOMB 2012 to be submitted to the Bioinformatics Journal and 2 Feng Zhao and Jinbo Xu Position specific distance dependent statistical potential for protein folding and assessment submitted to PNAS Computational prediction is important for proteins whose structure has not yet been or cannot be determined by experimental methods These largely unsolved problems are crucial for basic biological research and for applications in national priority areas such as pathogens biofuels remediation ecosystems metagenomics crops and nutrition top The Roux Group Benoit Roux University of Chicago with Jim Phillips of the UIUC NAMD group led work to port and tune NAMD for the Beagle XE6 system and its Gemini interconnect NAMD is one of the flagship workhorse applications of molecular dynamics for biological applications The results characterized Beagle s NAMD performance as excellent scaling comparing well to other leading machines like TACC s Lonestar completely obliterating older ones like Ranger and making our old rule of thumb of aiming for 1000 atoms core outdated The effort achieved multiple nanoseconds of simulated time day excellent performance for a general purpose NAMD This work benefits many applications as this code is widely used In addition the Roux Group with J C Grubart is using Beagle to perform molecular dynamics simulations to quantify the free energy of the insertion of membrane proteins and determine all of the contributions to that energy These results help to resolve a longstanding experiment simulation disagreement on the magnitude of the free energies for different amino acids This work has been submitted to Biophysical Journal as James Gumbart and Benoit Roux Determination of membrane insertion free energies by molecular dynamics simulations 2011 top Andrey Rzhetsky University of Chicago is developing tools to use Beagle for his project Assessing and Predicting Scientific Progress through Computational Language Understanding The project develops general methods relevant for students policy makers and scientists The goal is to generate dynamic high fidelity maps of knowledge claims in chemistry and related fields such as pharmaceuticals and toxicology Their initial use of Beagle involves optimization of parameters for graph based models of scientific collaboration This work has potential to help scientists gain knowledge in many health and non health domains particularly those where the literature has already grown to a size where human comprehension of all publications is no longer possible top Yves Lussier s group University of Chicago recently moved to UIC have been using Beagle to develop high throughput methods for identifying novel microRNA related targets for cancer therapy identifying the genetic underpinning of complex disease polymorphisms and computationally identifying novel unique personal variant polymorphisms associated to a disease They expect to complete a human phenome map of genome wide association studies GWAS used to identify common genetic factors that influence health and disease by the end of 2011 These studies are computationally intensive Part of this work was presented as Complex disease networks of trait associated SNPs unveiled by Information Theory at the AMIA Annual Symposium 2011 and will be included in JAMIA with a distinguished paper award Another paper is in preparation Mechanisms of Intergenic Polymorphisms and of Intragenic Pleiotropy of Complex Diseases Unveiled by Similarity and Overlap in eQTL Networks Complex Disease Gene Modules of A Cancer Associated SNPs Unveiled by Information Theory and Protein Interaction Networks Credit Lussier top The focus of the Hatsopoulos lab s Nicho Hatsopoulos University of Chicago research is the analysis of empirical neural activity based on Granger causal network models derived from generalized linear model of large scale networks These models are computationally intensive and are run efficiently as Matlab compiled executables on Beagle The group is starting to investigate how such networks of causally related neurons change over time Eventually when they have completed more efficient models they will shed light on how a group of neurons form a spatiotemporal pattern that is comparable with previously published local field potential spatiotemporal patterns and what the implication of those patterns are to behavior Results were presented at annual meeting of Society for Neuroscience Nov 2011 The team is currently developing C implementations of the estimation algorithms which will allow to scale up the models and test their performance on different Cray architectures such as the XK6 Example of causality networks estimated at different timings in relation to visual cue onset Before 101 50 ms At cue 51 200 ms After 201 350 ms Distribution of directions of LPF beta wave propagation during Cue period Credit Hatsopoulos top In non muscle cells contractile force production is governed by highly dynamic self organized actomyosin networks whose structure and mechanical properties emerge from a continuous interplay of assembly interaction and disassembly of actin filaments cross linkers and molecular motors To explore how Ed Munro and collaborators University of Chicago have developed agent based Brownian dynamics simulations that predicts macroscopic dynamics of contractility and stress relaxation based on purely local microscopic interactions among filaments cross linkers and motors Using Beagle they have studied the interplay of network architecture and turnover of network elements determine effective viscosity and network flow on long timescales and force dependent cross bridge kinetics and myosin minifilament assembly state control size and spacing of pulsed actomyosin contractions Two articles are in preparation by J Alberts and E M Munro and T Y Kim M Gardel and E M Munro Snapshot of agent based Brownian dynamics simulation of macroscopic dynamics of contractility and stress relaxation in actin filaments cross linkers and molecular motors Credit Munro top Epilepsy is a disease that is still poorly understood No obvious biological cause has been unequivocally identified and it appears to be a complex disease To understand its mechanics the Van Drongelen lab Wim van Drongelen University of Chicago has been working at performing realistic computational simulations that bridge the gap between the experimentally accessible individual behavior 100 cells and the clinically recorded aggregate behavior 1M They have ported their models to Beagle and have performed scaling and optimization studies which have been presented as Lee HC Hereld M Visser S Pesce L and van Drongelen W Scaling Behavior Of Computational Model of Neocortex poster at 4th International Meeting on Epilepsy Research Chicago IL May 19 21 2011 A paper describing those findings in more details will be prepared soon In the near future they plan to investigate the network and cell properties that appear to affect the dynamics of epileptic seizures through simulation of networks of neurons Parameter spaces will be studied to understand which appear to be the key factors A key part of the project is also to analyze experimental data from neurons grown in vitro Snapshot of neuronal activity in a simulation of brain neocortex Credit Van Drongelen top Kevin White and Bob Grossman University of Chicago are developing reliable and robust profiling and alignment software which requires regular and thorough testing almost every time a new version is produced This operation requires the reanalysis of a considerable amount of data In practice at this time it involves the regular analysis of a number of datasets currently available 2000 datasets Other tests might also be added in the future The scale of this operation requires a supercomputer such as Beagle to be able to provide a timely and reliable answer The team has developed workflows and successfully ported analysis tools and are now getting ready to perform a full scale test on Beagle top An important next step in host pathogen evolution studies is to incorporate models of within host dynamics into existing between host models which is being studied by Stefano Allesina and Greg Dwyer at the University of Chicago These within host models however are often difficult to incorporate because little is known about pathogen dynamics inside hosts As early as 1961 Saaty it was postulated that within host models could be compared using dose response time data yet the computational complexity involved prevented the implementation of this strategy Thus despite many within host models being proposed few have been used for parameter estimation and none have been tested Today with the advancement of computing technology in particular cluster computing it is possible to perform parameter estimation on mechanistic models of within host disease dynamics By implementing message passing on Beagle this team uses a simulation based approach for model selection on nested models to learn the relative importance of various mechanisms involved in virus growth in the gypsy moth caterpillar From this analysis they draw biological conclusions that provide the foundation to improve current practices in applying gypsy moth virus as an environmentally benign pesticide They also use their results to gain an understanding of host pathogen coevolution more generally To qualitatively assess the fit of the model to the data 104 parameter sets were generated from the joint posterior distribution of the parameters and used to simulate speed of kill The median 1st percentile and 99th percentile values for each time are plotted in red as a solid dotted line The overlaying black squares are the data collected during the dose response study Also contained in the figure is the virus dose received D and the number of larvae infected n Credit David A Kennedy top The Voth group Greg Voth University of Chicago has used Beagle to simulate actin which is the primary component of the eukaryotic cellular cytoskeleton and has been the target of much recent experimental work The specific focus of these simulations has been to characterize the structure and heterogeneity of the multi protein actin filament These simulations approach 1 million atoms in size and therefore the Beagle resource has been critical for achieving these research aims This work has contributed new knowledge on the behavior of actin filaments and also to our understanding of possible binding sites for therapeutic agents e g in cancer cells A snapshot from fully atomistic MD simulations of a 13 subunit piece of an actin filament used to understand the structural heterogeneity of this system and its response to nucleotide state Critical components to accurately simulate the filamentous state include the periodically repeating filament boundaries of the periodic cell are shown in blue and explicit solvation with appropriate ion concentrations The lower half of the simulation box shows the water environment potassium and chloride ions added to approximate physiological concentrations are shown as tan and cyan spheres respectively Credit Voth top The research group of Bob Eisenberg Rush University works on the calcium channels that control the contraction of the heart and the sodium channels that provide signaling in the nervous system They have systematically studied how these channels choose particular ions to perform their function In the time that Beagle has been operation it has allowed Eisenberg s group to make progress for example showing that dielectric boundaries have important roles in determining selectivity Improvement in channel function would have profound effects on the treatment of arrhythmias and diseases like multiple sclerosis that arise from poor function of these channels Ion Moving through a Bacterial Channel Credit Eisenberg top Non Health related projetcs Using the actor model of computation on Beagle Massively parallel applications are required to perform alignment free analyses of environmental soil microbiomes Data for such soil microbiomes are typically gathered by using shotgun DNA sequencing of all genomes found in such an environment called a metagenome Massively parallel tools are now required to tackle the big data in environmental research Pipelines exist for losely coupled tasks such as short read alignment but crafting tightly coupled genomic tools such as assemblers or alignment free approaches in general that can tackle huge datasets remains challenging Figure 1 A 512 node metagenome assembly job on Beagle The computation lasted 20 minutes 21 seconds The purple part at the beginning is input output This picture was generated with Spate a metagenome assembler in development and periodic patterns are visible Here we describe an ongoing project for building the next generation of tightly coupled high performance applications this is not targetting losely coupled tasks in genomics The idea of a Sequence Analysis Library SAL is that it can assist in solving various sequence based biology problems We have decided to use the actor model of computation 1 as the theoretical framework of computation We implemented a first principle object oriented distributed actor engine named Thorium This engine is designed and implemented for multi node multi core computations in mind although it also works on one node and or one core too Thorium also leverages mature and proven technologies such as C 1999 programming language MPI MPI 1 0 and Pthreads Figure 1 shows the basic design ideas of Thorium Our work is presently focused on the Spate application a metagenome assembler that is implemented strictly with actors Spate will allow us to tackle one of the U S Department of Energy DOE Joint Genome Institute JGI grand challenges namely the one entitled Great Prairie Soil Metagenome Grand Challenge This grand challenge contains 6 of soil metagenome samples from Kansas Iowa Wisconsin 1 cultivated corn soil sample and 1 native prairie soil sample for every of these 3 states All these samples have least roughly 2 billion DNA sequencing reads and the one with the higher number of reads has more than 5 billion reads 1 How did Beagle met our needs Spate is a metagenome assembler in development and it uses the Thorium actor model engine to drive the computation This project uses the C 1999 programming language We are currently testing the development code on Beagle using the DOE Great Prairie Soil Metagenome Grand Challenge 6 samples see ref 2 One of the sample is Iowa Continuous Corn Soil which contains 2 228 341 042 Illumina DNA sequences Using Beagle our goal is to obtain a metagenome assembly in less than 1 hour Right now it takes around 20 minutes 256 Beagle nodes to load data from storage to distributed memory and to build the assembly graph which contains 141 189 180 698 DNA sequences called k mers with 140 183 562 558 overlaps Beagle is a Cray XE6 product and has 726 nodes connected with a Cray Gemini interconnect This interconnect offers high throughput and low latency Each Beagle node has 2 AMD Opteron 6100 series processors each with 12 x86 64 cores total 24 cores and 32 GiB ECC DDR3 SDRAM of hysical memory This node architecture is suitable for our work because Thorium the underlying computation engine is a multi node multi core runtine environment and we typically use 1 single MPI rank and 22 or 23 worker threads per Beagle node The multi core processor AMD Opteron Magny Cours supports the x86 mfence instruction to allow us to insert memory barrier in the code to make sure that visibility of changes to memory is timely Submitting a job on Beagle is easy too We write a PBS script and submit it Also several C compilers are available on Beagle such as GNU gcc 4 8 1 and Cray craycc 8 1 4 Finally the amount of memory available per core 32 GiB 24 1 33 GiB on Beagle is a little higher than the amount of Mira at Argonne IBM Blue Gene Q has 1 GiB per core Beagles meets our needs by providing us with elastic high performance compute capability We are also thankful to

    Original URL path: http://beagle.ci.uchicago.edu/science-at-beagle/ (2014-11-13)
    Open archived version from archive

  • » Using Beagle
    your newl public key here in order to generate your SSH key Configuring Putty Start menu Pytty Putty In destination field enter login beagle ci uchicago edu Switch Protocol to SSH Type name of session in saved sessions Click Save Expand the SSH tab from the Category list Choose X11 from SSH list Check Enable X11 Forwarding Choose Auth from SSH list Upload your private key Choose Session from Category list Click Save Click Open Login with your key passphrase Close the shell Configuring Xming Go to http sourceforge net projects xming files latest download Install Xming Connecting Start Putty Start Xming Select your saved session Load your session Select Open Login in your shell You should now be able to run X applications like Matlab from the host on your local desktop top Guide for Optimal Performance on Beagle It is important to make a clear distinction between the resources requested i e what is specified in the PBS directives at the top of your job script and those actually used i e the values passed to the aprun flags n N d S For more details go to Basics about job submission on Beagle The PBS directives that appear at the top of a job script tell the job submission system how many compute nodes your job will require The options given to the aprun command determines the resources your job actually uses In order to make things clear we recommend that mppwidth is always the least upper bound number of processes and is a multiple of 24 that mppnppn is never used If compiler optimisation fails to improve performance significantly the next step is code optimisation Use CrayPAT to profile your code and understand the performance bottlenecks Hybrid MPI OpenMP programming may be an option to consider if each process in your code currently contains significant computational sections e g loops that may be more finely parallelised Hybrid codes make use of MPI for inter processor communication and OpenMP for intra process communication so that some or all cores within a node use multithreaded shared memory parallelism The advantage of this approach is that the threads within a process can share much of the same data structures thus providing more scope for cache optimisation and also reducing off node traffic See Hybrid MPI OpenMP script example It is possible to use all 24 cores within a node for shared memory OpenMP threads but this may not be desirable depending on how your code accesses memory See Examples of PBS scripts for Beagle for more information on these aspects of running jobs Note A single given compute node is always allocated to run a single user job multiple jobs never share a compute node If you want to run a couple of jobs on single nodes use our guide on Running Multiple Parallel Jobs Simultaneously or use Swift It is possible to write to files in parallel over multiple physical disks a technique called striping In striping your

    Original URL path: http://beagle.ci.uchicago.edu/using-beagle/ (2014-11-13)
    Open archived version from archive

  • » Software
    07 N A java PGI 3 1 49A 3 1 61 default N A PrgEnv pgi Python 2 7 1 default python 2 7 3 python 3 3 0 vanilla 2 7 3 vanilla gnu python R 2 12 1 2 14 1 default 2 15 1 gnu R top Debuggers Package Name Version Compiled With Module Name Abnormal Termination Processing ATP 1 1 0 1 1 1 1 1 2 1 3 0 default 1 5 2 N A atp Cray Performance Measurement Analysis Tools 5 1 3 default N A perftools GNU Debugger gdb 7 2 default cray gnu pgi gdb lgdb 1 2 1 4 default 1 5 cray gnu pgi xt lgdb Performance Application Programming Interface PAPI 4 1 0 0 2 Default 4 1 0 cray gnu pgi xt papi STAT 1 1 3 default 1 1 2 1 2 1 3 N A stat top Libraries Package Name Version Compiled With Module Name AMD Core Math Library ACML 4 4 0 default 5 1 0 gnu pgi acml Fastmv 1 0 8 1 0 9 default cray gnu pgi libfast FFTW2 2 1 5 2 cray gnu pgi fftw 2 1 5 2 FFTW3 2 1 5 2 2 1 5 4 3 2 2 1 default 3 3 0 1 cray gnu pgi fftw Global Arrays GA 4 3 2 4 3 3 4 3 5 default 5 0 2 cray gnu pgi ga GSL 1 14 default 1 15 cray gnu pgi gsl HDF5 1 8 5 0 cray gnu pgi hdf5 HDF5 Parallel 1 8 5 0 cray gnu pgi hdf5 parallel IOBUF 2 0 1 2 0 2 default cray gnu pgi iobuf libonesided 1 0 0 1 1 0 1 3 0 default 1 5 0 cray gnu pgi onesided LibSci 10 4 9 10 5 0 10 5 02 11 0 01 default 11 1 01 cray gnu pgi xt libsci METIS 4 0 1 default gnu metis MRNet 3 0 0 default cray gnu pgi mrnet NetCDF 4 1 1 0 cray gnu pgi netcdf NetCDF Parallel 4 1 1 0 cray gnu pgi netcdf hdf5parallel Netlib CBLAS 3 0 default gnu pgi cblas NUMA Toolkit 1 0 0 1 1 0 1 3 0 default 1 5 0 cray gnu pgi ntk PETSc 3 1 04 3 1 05 3 1 08 cray gnu pgi petsc PETSc Complex Numbers 3 1 04 3 1 05 3 1 08 cray gnu pgi petsc complex Qhull 2010 1 default gnu qhull qrupdate 1 1 1 default gnu qrupdate SuiteSparse 3 6 0 default gnu SuiteSparse Tcl 8 5 7 default cray gnu pgi tcl top Python Modules Package Name Version Compiled With Module Name NumPy 1 5 1 python2 7 numpy NumPy 1 7 0 python2 7 export path Python Imaging Library PIL 1 1 7 python2 7 pil SciPy 0 9 0rc3 python2 7 scipy SciPy 0 12 0 python2 7 export path Matplotlib 1 2 1

    Original URL path: http://beagle.ci.uchicago.edu/software/ (2014-11-13)
    Open archived version from archive

  • » Technical Specification
    two cores connected to the same Gemini chip plus a little over 0 1 μs for each Gemini chip that is between the sender and the receiver node Each Gemini chip has 168 GB s bandwidth of switching capacity Each Gemini Chip has 48 ports Injection bandwidth per node is 20 GB s The Gemini chips are arranged on a 3 dimensional torus with 10 links from each router on to the high performance network The peak bi directional bandwidth of each link is 8 GB s and the latency is around 1 1 5μs MPI 1 2 μs latency 15 Million independent messages per second per NIC two NICs per Gemini PGAS support In addition to supporting MPI over the standard programming languages of C C and Fortran the Gemini interconnect has direct hardware support for partitioned global address space PGAS programming models including Unified Parallel C UPC Co array Fortran and Chapel Gemini allows remote references to be pipelined in these programming models which can result in orders of magnitude performance improvement over library based message passing models top Storage Infiniband connected The Lustre filesystem data is stored on a DDN 10000 storage array consisting of ten 60 disk capacity enclosures connected via Infiniband to eight dedicated Cray XIO service nodes The enclosures are 50 filled with 2 TB drives providing 600TB raw storage subsequently configured as RAID 6 logical volumes The Lustre filesystem metadata is stored on a Fiber Channel storage array connected to a dedicated XIO service node Lustre file system Users have to create their own directory on Lustre Compute nodes only have access to Lustre lustre beagle not your home directory Files which are read or written by compute nodes must be on Lustre Design your applications to use fewer but larger files LUSTRE filesystem prefers to handle a small number of large files No long term storage of data It s a scratch space Not backed up In the case of a hardware or software failure we can not guarantee that data will not be lost Lustre allows users to control the striping parameters when storing data on the filesystem top Network Beagle has a High Speed Network HSN with a 10 Gb connection to the internet specifically to the Argonne Mathematics and Computer Science MCS Division s HPC switch which has 10 Gb connectivity to MREN Metropolitan Research Education Network and ESNet Energy Sciences Network The University of Chicago campus is also connected at 10 Gb to MREN top Operating System Beagle uses Cray Linux Environment v3 CLE3 a suite of high performance software including a SUSE Linux based operating system designed to run large complex applications The Linux environment features Compute Node Linux CNL as the default compute kernel When running highly scalable applications CNL runs in Extreme Scalability Mode ESM which ensures operating system services do not interfere with application scalability Real world applications have proven that this scales to more than 200 000 cores Users can also run industry

    Original URL path: http://beagle.ci.uchicago.edu/technical-specification/ (2014-11-13)
    Open archived version from archive

  • » News
    Cloud using the potential of cloud computing to bring advanced scientific computing infrastructure to researchers around the world June The first 40 fellows of the Data Science for Social Good Summer Fellowship arrived in Chicago for 12 busy weeks developing solutions for non profits and government agencies DSSG Director Rayid Ghani spoke at Techweek Chicago about how the program offers a new path for students interested in using technology to improve the world Computation Institute researchers were well represented at the University s Alumni Weekend with Senior Fellow Gary An giving a talk about modeling in biomedicine UrbanCCD director Charlie Catlett appearing on an urban data panel with Harris School of Public Policy Dean Colm O Muircheartaigh and Lewis Sebring Distinguished Service Professor Stephen W Raudenbush and Director Ian Foster talking about how Big Computation can unlock bigger knowledge July Argonne National Laboratory with help from Sen Dick Durbin officially unveiled their newest supercomputer Mira a 10 petaflop IBM Blue Gene Q machine that ranks among the top ten fastest computers in the world CI Senior Fellow Benoit Roux published new findings in the journal Nature on how water affects one of the most important proteins for life the potassium channel August To address the challenges of planning the massive 600 acre Chicago Lakeside Development on the South Side UrbanCCD researchers linked up with developers McCaffery Interests and architects at Skidmore Owings Merrill for the LakeSim project a computational platform for large scale urban design In a paper for Cancer Research CI Fellow and Faculty Samuel Volchenboum and CI Fellow Mike Wilde published a new algorithm to help researchers find better genetic classifiers for diagnosing and treating cancer The inaugural class of the Data Science for Social Good Fellowship summarized their summer work in a Data Slam event held downtown at the University of Chicago Gleacher Center September The Knowledge Lab s Metaknowledge Research Network held their first meeting in Pacific Grove California strategizing their approach to questions such as What makes someone a great scientist or inventor A team led by CI Senior Fellow and Faculty Andrey Rzhetsky analyzed over 120 million patient records and thousands of scientific studies to create a groundbreaking genetic map of complex diseases research published in the journal Cell October No news was bigger this year than the naming of CI Senior Fellow Lars Peter Hansen as one of three recipients of the 2013 Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel Hansen received the honor along with UChicago colleague Eugene Fama and Yale s Robert Shiller and was credited with important statistical work in the field of econometrics A panel of UChicago economists later celebrated the significance of Hansen and Fama s contributions to the field and our knowledge about the behavior of financial markets Other Nobel Prizes awarded this month acknowledged computational work in discovering the Higgs Boson and studying chemical changes The Texas Advanced Computing Center announced Wrangler a new open source data analysis and management system that will use Globus services for data transfer The Center for Robust Decision Making on Climate and Energy Policy held their annual all hands meeting in Chicago and discussed research that is improving the accuracy and scope of computer models for climate agriculture and the economy As scientific instruments collect more data and more complex data their computational demands soar CI Senior Fellow Andrew Chien launched a project with measurement company Agilent to find new pattern matching strategies for analyzing that data more quickly and efficiently In a paper published by PLoS Genetics CI Senior Fellow Nancy Cox addressed one of the most important mysteries in the study of genetics and disease the case of missing heritability UrbanCCD Director Charlie Catlett spoke to an architecture conference about the potential of data to transform how cities are designed and built November The Computation Institute invited the University of Chicago campus to a series of lightning talks short presentations of ongoing research and opportunities for student involvement Watch videos from the event at the CI YouTube channel The Social Sciences Division produced a video and feature on CI Senior Fellow James Evans and his new Knowledge Lab Evans also wrote an editorial in Science discussing how modeling scientific impact and using robot scientists to suggest experiments could revolutionize research A panel of CI and UrbanCCD experts discussed the data based future of city planning and development in a special UChicago Discovery Series event Chicago City of Big Data CI Director Ian Foster and Fellow Ravi Madduri presented the discovery accelerating powers of Globus and Globus Genomics at Amazon Web Services re Invent conference December A multi institutional collaboration including the CI announced the new Center for Hierarchical Materials Design to explore innovative new avenues in the creation of materials for technology medicine and energy As part of a special section in the Proceedings of the National Academy of Sciences CI Fellow Joshua Elliott and other RDCEP researchers published a study modeling the effect of climate change on the world s freshwater supply and agriculture PATRIC the world s largest database of genomic information about pathogenic bacteria prepared for its tenth anniversary by surpassing 10 000 annotated genomes University of Chicago students tested the limits of the new Hack Arts Lab in the first ever digital fabrication course WRITTEN BY ROB MITCHUM To read this article in original please visit this link Lower priority jobs For the new low priority settings the restrictions are a walltime of four hours or less and nothing more than 50 nodes Basically we re scheduling normal priority jobs first then filling in the gaps with lower priority jobs The smaller the job the easier it ll be to schedule so the quicker it will move from queued to running Nothing needs to be specified in the submit script Low priority configurations are handled on the server side on a per project basis Any project that has been designated as having a low priority allocation will automatically have those

    Original URL path: http://beagle.ci.uchicago.edu/news-and-events/ (2014-11-13)
    Open archived version from archive



  •