Oser Communications Group

Super Computer Show Daily Nov 20, 2014

Issue link: http://osercommunicationsgroup.uberflip.com/i/412890

Contents of this Issue


Page 3 of 11

S u p e r C o m p u te r S h o w D a i l y Th u r s d a y, N o ve m b e r 2 0 , 2 0 1 4 4 By Bruce McCormick, Cognimem Founder A rapidly emerging theme in large-scale data mining is the use of models based on network theory, also known as graph the- ory, which allow abstract representation of associations (or "connections") between entities (or "nodes"). The power of this approach mostly lies in the fact that very large amounts of data can be reduced to a limited number of parame- ters per node, representing the "role" that each node has in the network. A truly remarkable fact is that very different sys- tems, such as social networks, brain net- works and networks representing interac- tions among financial entities, genes or logistics hubs, tend to share some archi- tectural properties. The most prominent of these properties is the presence of "hubs," i.e. a small proportion of nodes that have a very high number of connec- tions with other entities across the net- work: for example, social leaders, key areas in the brain cortex, commodity funds and international airports. Finding hub nodes can be a highly computationally demanding operation, as it requires measuring the intensity of connection between all pairs of nodes; in a network of n nodes, n^2-n opera- tions are required. While the definition of connectivity varies among fields, there is growing consensus that basic norms, in spite of (or thanks to) their simplicity often repre- sent an acceptable approximation to much more complex formulae, subject to appropriate data transformation. L1- norm is well-suited to represent correla- tion and synchronization between time- series, L0-norm is appropriate for com- paring symbol strings, as found in text or gene sequences, and the Lsup-norm is relevant to situations where emphasis has to be given to mismatch even in a single parameter, for example when comparing spectra or visual features. Silicon from CogniMem Inc. that natively implements the nonlinear clas- sifiers kNN and RBF in hardware (based on the L1 norm), represents an ideal platform for solving some of these prob- lems. It allows massively-parallel calcu- lation of norms between a stored set of reference vectors and vectors that broadcast simultaneously to all process- ing units; as such, it can be used to determine connectedness in networks of arbitrary size. A crucial feature is that results are sorted on-the-fly as they are read out, enabling rapid identification of the nodes having lowest norms without necessitating download of all results from all processing units. CogniMem's current flagship prod- uct, the CM1K, harbors 1,024 process- ing nodes running at up to 27 MHz, and is up to 90 times faster than a DSP at 300 MHz in comparing a 256-byte pat- tern to 1,024 stored patterns. Small clus- ters of paralleled CM1Ks can easily exceed the performance of current multi-GHz CPUs, at a fraction of the power. A new chip, expected for release in 2015, will add another ten-fold increase to processing power. Come hear our presentation at the Exhibitor's Forum on Thursday afternoon. Visit Cognimem Technologies at booth #3744. For more information, visit www.cognimem.com, call 916-358-9485 or email info@cognimem.com. USING SIMPLE NORMS TO APPROXIMATE, ACCELERATE COMPLEX CALCULATIONS NUMAQ – A SCALABLE IN-MEMORY PROCESSING AND BIG DATA ANALYTICS APPLIANCE By Trond Smestad, President, Numascale LLC NumaQ represents a new generation of in-memory-based analytics appliances that scales to thousands of cores and terabytes of memory for data and memory-intensive workloads. Numascale offers an innovative approach to big data analytics by pro- viding in-memory capabilities via NumaQ systems. NumaQ, developed by Numascale's department in Singapore, employs Numascale's scal- able shared memory technology to offer a new generation of high per- formance in-memory analytics capa- bilities that provide the power of a cluster coupled with the simplicity of a desktop. NumaQ is a state-of-the-art x86 shared memory system from 1degreenorth that is specifically designed for big data in-memory ana- lytics. The system uses high quality servers, connected with Numascale's NumaConnect scalable SMP fabric and a tuned Linux operating system. 1degreenorth's engineers have inte- grated the popular open source R sta- tistical programming environment from Revolution Analytics with Numascale's NumaManager to provide a desktop-like experience in a com- modity cluster system, with hundreds of cores and terabytes of RAM that is easy to manage and use. Designed and specifically engi- neered for in-memory big data analyt- ics, NumaQ accelerates processing by providing a set of optimized appli- ances. The in-memory analytics appli- ance is supported by the R2 analytics programming environment and the MonetDB column-store analytics data- base. In addition, an in-memory Apache Spark Appliance is provided by the R2 analytics programming envi- ronment, just as with Hadoop clusters. The NumaQ standard system con- figurations scale from 128 cores with 1TB memory to 4,096 cores with 32TB of global shared memory in a single system image environment. The sys- tems are based on Dell's industry-stan- dard server platform, Red Hat ® Enterprise Linux ® and Numascale's NumaConnect. Numascale has also developed the technology that enables NumaQ. The NumaConnect products turns a collec- tion of standard servers with separate memories and I/O into a unified sys- tem that delivers the functionality of high-end enterprise servers and main- frames at a fraction of the cost. NumaQ, as an appliance, turns a collection of standard servers with sep- arate memories and I/O into a unified system that delivers the functionality of high-end enterprise servers and mainframes at a fraction of the cost. NumaConnect links commodity servers together to form a single uni- fied system where all processors can coherently access and share all memo- ry and I/O. The combined system runs a single instance of a standard Linux OS. The result is an affordable shared memory computing option to tackle data-intensive applications. NumaConnect-based systems running with entire data sets in memory are "orders of magnitude faster than clus- ters or systems based on any form of existing mass-storage devices and will enable data analysis and decision sup- port applications to be applied in new and innovative ways," said Laurence Liew, Numascale Vice President. NumaConnect provides an afford- able solution by delivering all the advantages of expensive shared memo- ry computing – streamlined application development, the ability to compute on large datasets, the ability to run more rigorous algorithms, enhanced scala- bility and more – for a cluster price point. Visit Numascale at booth #1923. For more information, go to www .numascale.com, call 832-470-8200 or email ts@numascale.com. OFFERED FOR YOUR PREPRANDIAL EDIFICATION David Abramson of the University of Queensland will discuss some of the vex- ing problems of debugging software in software running on supercomputers that exploit massive parallelism in a talk titled "It Was Working Until I Changed...," scheduled for 11:15 a.m. to noon on Tuesday, November 18 in the New Orleans Theater. According to his abstract for the talk, "This process becomes even more difficult in supercomputers that exploit massive parallelism because the state is distributed across processors, and additional failure modes (such as race and timing errors) can occur." Abramson will discuss a debugging strategy called "relative debugging," which allows a user to compare the run time state between executing programs, one being a working, "reference" code, and the other being a test version. His discussion will review the basic ideas in relative debugging and will give exam- ples of how it can be applied to debug- ging supercomputing applications. If that's not your cup of tea, you could also head down to room 393-395 at 11:30 to hear a finalist for the conference's best student paper, "A Volume Integral Equation Stokes Solver for Problems with Variable Coefficients," presented by authors Dhairya Malhotra, Amir Gholami, George Biros from the University of Texas at Austin. According to their abstract, the paper presents a novel numerical scheme for solving the Stokes equation with vari- able coefficients in the unit box that's based on a volume integral equation for- mulation. The session will be chaired by Justin Luitjens of NVIDIA. At the same time next door in room 391-392, Suzanne Rivoire from Sonoma State University will be chairing a ses- sion in which authors Niladrish Chatterjee, Mike O'Connor, Gabriel H. Loh, Nuwan Jayasena and Rajeev Balasubramonian offer their paper titled "Managing DRAM Latency Divergence in Irregular GPGPU Applications." The authors will propose solutions for this problem that yield a 10.1 percent per- formance improvement for irregular GPGPU workloads relative to a through- put-optimized GPU memory controller. AN INDEPENDENT PUBLICATION NOT AFFILIATED WITH SC Lee M. Oser CEO and Editor-in-Chief Kim Forrester Paul Harris Associate Publishers Lorrie Baumann Editorial Director Jeanie Catron JoEllen Lowry Associate Editors Yasmine Brown Vicky Glover Graphic Designers Mary Procida Caitlyn Roach Customer Service Managers Enrico Cecchi European Sales Super Computer Show Daily is published by Oser Communications Group ©2014 All rights reserved. Executive and editorial offices at: 1877 N. Kolb Road • Tucson, AZ 85715 520.721.1300 • Fax: 520.721.6300 www.oser.com European offices located at Lungarno Benvenuto Cellini, 11, 50125 Florence, Italy.

Articles in this issue

Links on this page

view archives of Oser Communications Group - Super Computer Show Daily Nov 20, 2014