BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/Chicago X-LIC-LOCATION:America/Chicago BEGIN:DAYLIGHT TZOFFSETFROM:-0600 TZOFFSETTO:-0500 TZNAME:CDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0500 TZOFFSETTO:-0600 TZNAME:CST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260522T150117Z LOCATION:D161 DTSTART;TZID=America/Chicago:20181112T090000 DTEND;TZID=America/Chicago:20181112T173000 UID:submissions.supercomputing.org_SC18_sess158@linklings.com SUMMARY:9th Workshop on Latest Advances in Scalable Algorithms for Large-S cale Systems DESCRIPTION:Novel scalable scientific algorithms are needed in order to en able key science applications to exploit the computational power of large- scale systems. This is especially true for the current tier of leading pet ascale machines and the road to exascale computing as HPC systems continue to scale up in compute node and processor core count. These extreme-scale systems require novel scientific algorithms to hide network and memory la tency, have very high computation/communication overlap, have minimal comm unication, and have no synchronization points. With the advent of Big Data in the past few years the need of such scalable mathematical methods and algorithms able to handle data and compute intensive applications at scale becomes even more important. \n\nScientific algorithms for multi-petaflop and exaflop systems also need to be fault tolerant and fault resilient, s ince the probability of faults increases with scale. Resilience at the sys tem software and at the algorithmic level is needed as a crosscutting effo rt. Finally, with the advent of heterogeneous compute nodes that employ st andard processors as well as GPGPUs, scientific algorithms need to match t hese architectures to extract the most performance. This includes differen t system-specific levels of parallelism as well as co-scheduling of comput ation. Key science applications require novel mathematics and mathematical models and system software that address the scalability and resilience ch allenges of current- and future-generation extreme-scale HPC systems.\n\nC ommunication Avoiding Multigrid Preconditioned Conjugate Gradient Method f or Extreme Scale Multiphase CFD Simulations\n\nA communication avoiding (C A) multigrid preconditioned conjugate gradient method (CAMGCG) is applied to the pressure Poisson equation in a multiphase CFD code JUPITER, and it s computational performance and convergence property are compared against CA Krylov methods. A new geometric multigrid precon...\n\n\nYasuhiro Idomu ra, Takuya Ina, Susumu Yamashita, Naoyuki Onodera, and Susumu Yamada (Japa n Atomic Energy Agency) and Toshiyuki Imamura (RIKEN)\n------------------- --\nShift-Collapse Acceleration of Generalized Polarizable Reactive Molecu lar Dynamics for Machine Learning-Assisted Computational Synthesis of Laye red Materials\n\nReactive molecular dynamics is a powerful simulation meth od for describing chemical reactions. Here, we introduce a new generalized polarizable reactive force-field (ReaxPQ+) model to significantly improve the accuracy by accommodating the reorganization of surrounding media. Th e increased computati...\n\n\nKuang Liu, Subodh Tiwari, Chunyang Sheng, Ar avind Krishnamoorthy, Sungwook Hong, Pankaj Rajak, Rajiv K. Kalia, Aiichir o Nakano, Ken-ichi Nomura, and Priya Vashishta (University of Southern Cal ifornia); Manaschai Kunaseth (National Science and Technology Development Agency, Thailand); Saber Naserifar and William A. Goddard III (California Institute of Technology); Ye Luo and Nichols A. Romero (Argonne National L aboratory); and Fuyuki Shimojo (Kumamoto University)\n-------------------- -\nWorkshop Lunch (on your own)\n---------------------\nWorkshop Afternoon Break\n---------------------\nA General-Purpose Hierarchical Mesh Partiti oning Method with Node Balancing Strategies for Large-Scale Numerical Simu lations\n\nLarge-scale parallel numerical simulations are essential for a wide range of engineering problems\n that involve complex, coupled physic al processes interacting across a broad range of spatial\n and temporal s cales. The data structures involved in such simulations (meshes, sparse ma trices, etc.) are...\n\n\nFande Kong (Idaho National Laboratory); Roy H. S togner (University of Texas); and Derek R. Gaston, John W. Peterson, Cody J. Permann, Andrew E. Slaughter, and Richard C. Martineau (Idaho National Laboratory)\n---------------------\nKeynote 3: Hierarchical Algorithms on Hierarchical Architectures\n\nSome algorithms achieve optimal arithmetic c omplexity with low arithmetic intensity (flops/Byte), or possess high arit hmetic intensity but lack optimal complexity, while some hierarchical algo rithms, such as Fast Multipole and its H-matrix algebraic generalizations, realize a combination of optimal ...\n\n\nDavid Keyes (King Abdullah Univ ersity of Science and Technology)\n---------------------\nIntroduction\n\n Vassil Alexandrov (Barcelona Supercomputing Center)\n--------------------- \nKeynote 1: A Few Scheduling Problems for Resilience at Scale\n\nThe talk will address scheduling problems related to multi-level checkpointing, IO interference,\nsilent errors detection and correction, and workflows.\n\n \nYves Robert (ENS Lyon)\n---------------------\nLow Thread-Count Gustavso n: A Multithreaded Algorithm for Sparse Matrix-Matrix Multiplication Using Perfect Hashing\n\nSparse matrix-matrix multiplication is a critical kern el for several scientific computing applications, especially the setup pha se of algebraic multigrid. The MPI+X programming model, which is growing i n popularity, requires that such kernels be implemented in a way that expl oits on-node parallelism...\n\n\nJames Elliott and Christopher Siefert (Sa ndia National Laboratories)\n---------------------\nIterative Randomized A lgorithms for Low Rank Approximation of Terascale Matrices with Small Spec tral Gaps\n\nRandomized approaches for low rank matrix approximations have become popular in recent years and often offer significant advantages ove r classical algorithms because of their scalability and numerical robustne ss on distributed memory platforms. We present a distributed implementatio n of randomized b...\n\n\nChander J. Iyer (Rensselaer Polytechnic Institut e, Yahoo! Research); Alex Gittens and Christopher D. Carothers (Rensselaer Polytechnic Institute); and Petros Drineas (Purdue University)\n--------- ------------\nKeynote 2: HPC and AI as Drivers for Industrial Engagement\n \nThe Hartree Centre is transforming UK industry through high performance computing, big data and cognitive technologies. Backed by over £170 millio n of government funding and significant strategic partnerships with organi sations such as IBM and Atos, the Hartree Centre is home to some of the mo st tec...\n\n\nAlison Kennedy (Hartree Centre)\n---------------------\nWor kshop Morning Break\n---------------------\nMachine Learning-Aided Numeric al Linear Algebra: Convolutional Neural Networks for the Efficient Precon ditioner Generation\n\nMarkus Götz received his Bachelors and Masters degr ee in Software Engineering from the University of Potsdam in 2010 and 2014 respectively. Afterwards, he has been with the Research Center Jülich and the University of Iceland, from which Markus obtained his PhD degree in C omputational Engineering fo...\n\n\nMarkus Götz (Karlsruhe Institute of Te chnology)\n---------------------\nIntroduction - 9th Workshop on Latest Ad vances in Scalable Algorithms for Large-Scale Systems\n\nNovel scalable sc ientific algorithms are needed in order to enable key science applications to exploit the computational power of large-scale systems. This is especi ally true for the current tier of leading petascale machines and the road to exascale computing as HPC systems continue to scale up in ...\n\n\nVass il Alexandrov (Barcelona Supercomputing Center), Al Geist (Oak Ridge Natio nal Laboratory), Jack Dongarra (University of Tennessee), and Christian En gelmann (Oak Ridge National Laboratory)\n---------------------\nEvent-Trig gered Communication in Parallel Computing\n\nCommunication overhead in par allel systems can be a significant bottleneck in scaling up parallel compu tation. In this paper, we propose event-triggered communication methods to reduce such communication overhead for numerical simulation of partial di fferential equations. As opposed to traditional c...\n\n\nSoumyadip Ghosh, Kamal Saha, and Vijay Gupta (University of Notre Dame) and Gretar Tryggva son (Johns Hopkins University)\n---------------------\nOn Advanced Monte C arlo Methods for Linear Algebra on Advanced Accelerator Architectures\n\nI n this paper we present computational experiments performed using the Mark ov Chain Monte Carlo Matrix Inversion (MCMCMI) on several architectures of NVIDIA accelerators and two iterations of the Intel x86 architecture and investigate their impact on performance and scalability of the method.\nTh e me...\n\n\nAnton Lebedev (University of Tübingen) and Vassil Alexandrov (ICREA, Barcelona Supercomputing Center)\n---------------------\nNon-Colle ctive Scalable Global Network Based on Local Communications\n\nTo efficien tly perform collective communications in current high-performance computin g systems is a time-consuming task.\nWith future exascale systems, this co mmunication time will be increased further.\nHowever, global information i s frequently required in various physical models.\nBy exploiting domai...\ n\n\nMarco Berghoff and Ivan Kondov (Karlsruhe Institute of Technology)\n- --------------------\nCommunication Reduced Multi-Timestep Algorithm for R eal-Time Wind Simulation on GPU-Based Supercomputers\n\nWe develop a commu nication reduced multi-time-step (CRMT) algorithm for a Lattice Boltzmann method (LBM) based on a block-structured adaptive mesh refinement (AMR). T his algorithm is based on the temporal blocking method, and can improve co mputational efficiency by replacing a communication bottlene...\n\n\nNaoyu ki Onodera, Yasuhiro Idomura, and Yussuf Ali (Japan Atomic Energy Agency) and Takashi Shimokawabe (University of Tokyo)\n---------------------\nDyna mic Load Balancing of Plasma and Flow Simulations\n\nExtracting performanc e from simulations with complex information dependencies\n on massively p arallel computers requires the computational work to be evenly\n distribu ted across the processing resources while maintaining low\n communication costs.\n Plasma simulations using a particle-in-cell method...\n\n\nGerr ett Diamond and Cameron W. Smith (Rensselaer Polytechnic Institute), Eisun g Yoon (Ulsan National Institute of Science and Technology), and Mark S. S hephard (Rensselaer Polytechnic Institute)\n\nTag: Algorithms, Heterogeneo us Systems, Resiliency\n\nRegistration Category: Workshop Reg Pass\n\nSess ion Chairs: Vassil Alexandrov (Hartree Centre, STFC); Jack Dongarra (Unive rsity of Tennessee, Knoxville; Oak Ridge National Laboratory (ORNL)); Chri stian Engelmann (Oak Ridge National Laboratory (ORNL)); and Al Geist (Oak Ridge National Laboratory (ORNL)) END:VEVENT END:VCALENDAR