BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20250626T234154Z LOCATION:B310 DTSTART;TZID=America/New_York:20241117T090000 DTEND;TZID=America/New_York:20241117T173000 UID:submissions.supercomputing.org_SC24_sess737@linklings.com SUMMARY:IA^3 2024 - 14th Workshop on Irregular Applications: Architectures & Algorithms DESCRIPTION:Due to the heterogeneous datasets they process, data-intensive applications employ diverse methods and data structures, exhibiting irreg ular data accesses, control flows, and communication patterns. Modern data analytics applications additionally require supporting dynamic data struc tures, asynchronous control flows, and mixed parallel programming models. Supercomputing systems are organized around software and hardware optimize d for data locality and bulk synchronous computations. Managing irregular behaviors requires a substantial programming effort and lacks integration, leading to poor performance. Holistic solutions to these challenges emerg e only by considering the problem from multiple perspectives: from micro- to system-architectures, from compilers to languages, from libraries to ru ntimes, and from algorithm design to data characteristics. Only collaborat ive efforts among researchers with different expertise, including domain e xperts and end-users, can lead to significant breakthroughs. This workshop brings together scientists with different backgrounds to discuss methods and technologies for efficiently supporting irregular applications on curr ent and future architectures.\n\nEstablish the basis for Breadth-First Sea rch on Frontier System: XBFS on AMD GPUs\n\nGraphics Processing Units (GPU s) offer significant potential for accelerating various computational task s, including Breadth-First Search (BFS). Numerous efforts have been made t o deploy BFS on GPUs effectively. To address the dynamic nature of BFS, XB FS, the state-of-the-art work, employs an adapti...\n\n\nHaoshen Yang (Rut gers University), Hao Lu and Naw Safrin Sattar (Oak Ridge National Laborat ory (ORNL)), Hang Liu (Rutgers University), and Feiyi Wang (Oak Ridge Nati onal Laboratory (ORNL))\n---------------------\nIA^3 — Lunch Break\n------ ---------------\nInvited Talk: Discussion on Hash-Table Approaches for Eff icient Sparse Tensor Contraction\n\nSparse tensor contraction (SpTC) is a crucial operation in high-performance applications, particularly in comput ational chemistry, high-order tensor decompositions, and quantum sciences. This talk will explore the performance challenges associated with SpTC an d review current state-of-the-art soluti...\n\n\nJiajia Li (North Carolina State University)\n---------------------\nGPU Accelerated Sparse Cholesky Factorization\n\nThe solution of sparse symmetric positive definite linea r systems is an important computational kernel in large-scale scientific a nd engineering modeling and simulation. We will solve the linear systems u sing a direct method, in which a Cholesky factorization of the coefficient matrix is performed u...\n\n\nM. Ozan Karsavuran and Esmond G. Ng (Lawren ce Berkeley National Laboratory (LBNL)) and Barry W. Peyton (Dalton State College)\n---------------------\nIA^3 — Morning Break\n------------------- --\nEnhancing Small Message Aggregation with Directive-Based Deferred Exec ution\n\nThe partitioned global address space (PGAS) model offers one-side d communication operations to efficiently access local and remote data thr ough a distributed shared memory model using point-to-point network operat ions. An extension to the OpenSHMEM PGAS library previously demonstrated how message a...\n\n\nAaron Welch and Oscar Hernandez (Oak Ridge National Laboratory (ORNL)) and Stephen Poole and Wendy Poole (Los Alamos National Laboratory (LANL))\n---------------------\nEnhancing Scalability and Perfo rmance in Influence Maximization with Optimized Parallel Processing\n\nInf luence Maximization (IM) is vital in viral marketing and biological networ k analysis for identifying key influencers. Given its NP-hard nature, appr oximate solutions are employed. This paper addresses scalability challenge s in a scale-out shared memory system, by focusing on the state-of-the-art ...\n\n\nHanjiang Wu, Huan Xu, and Joongun Park (Georgia Institute of Tec hnology); Jesmin Jahan Tithi, Fabio Checconi, Jordi Wolfson-Pou, and Fabri zio Petrini (Intel Corporation); and Tushar Krishna (Georgia Institute of Technology)\n---------------------\nPredicting Compute Node Unavailability in HPC: A Graph-Based Machine Learning Approach\n\nAs high-performance co mputing (HPC) systems advance towards Exascale computing, their size and c omplexity increase, introducing new maintenance challenges. Modern HPC sys tems feature data monitoring infrastructures that provide insights into th e system's state. This data can be leveraged to train ma...\n\n\nJože Roža nec (Jožef Stefan Institute)\n---------------------\nPerformance evaluatio n and modelling of single-precision matrix multiplication on Cerebras CS-2 \n\nAlthough recent supercomputers have been improving their computational performance, achieving performance scaling with respect to the number of nodes is not easy due to long inter-node communication latency. Many attem pts have been made to hide communication latency and maintain strong scala bility e...\n\n\nRyunosuke Matsuzaki (Meiji University), Daichi Mukunoki ( Independent), and Takaaki Miyajima (Meiji University)\n------------------- --\nInvited Talk: Accelerating Irregular Algorithms on GPUs\n\nMartin Burt scher (Texas State University)\n---------------------\nShared Memory-Aware Latency-Sensitive Message Aggregation for Fine-Grained Communication\n\nM essage aggregation is widely used with a goal to reduce communication cost in HPC applications. The discrepancy in the order of overhead of sending a message and cost of per byte transferred motivates the need for message aggregation, for several irregular fine-grained messaging applications lik e g...\n\n\nKavitha Chandrasekar (University of Illinois) and Laxmikant Ka le (University of Illinois Urbana-Champaign)\n---------------------\nLinea r Algebra Approach for Directed Triad Counting and Enumeration\n\nTriangle counting and enumeration are commonly used in real-world applications on directed graphs. However, the performance of triangle counting algorithms is usually benchmarked on undirected graphs. As such, many of these algori thms and formulations are not suitable for identifying the types of di...\ n\n\nYuttapichai Kerdcharoen (Carnegie Mellon University, CMKL University) ; Upasana Sridhar (Carnegie Mellon University); Orathai Sangpetch (CMKL Un iversity); and Tze Meng Low (Carnegie Mellon University)\n---------------- -----\nConcluding Remarks\n\nAntonino Tumeo and John Feo (Pacific Northwes t National Laboratory (PNNL)); Michela Becchi (North Carolina State Univer sity); and Ana Lucia Verbanescu (University of Twente, Netherlands)\n----- ----------------\nIA ^3 Debate\n\nAntonino Tumeo (Pacific Northwest Nation al Laboratory (PNNL)), Ariful Azad (Indiana University), Giulia Guidi (Cor nell University), John Leidel (Tactical Computing Laboratories LLC), and G eorgios Michelogiannakis (Lawrence Berkeley National Laboratory)\n-------- -------------\nEfficient Tree-based Parallel Algorithms for N-Body Simulat ions Using C++ Standard Parallelism\n\nThe Barnes-Hut approximation for N- body simulations reduces the time complexity of the naive all-pairs approa ch from O(N^2) to O(N log N) by hierarchically aggregating nearby particle s into single entities using a tree data structure. \nThis inherently irre gular algorithm poses substantial challenges...\n\n\nThomas Lane Cassell a nd Tom Deakin (University of Bristol, England); Aksel Alpay and Vincent He uveline (Heidelberg University); and Gonzalo Brito Gadeschi (NVIDIA Corpor ation)\n---------------------\nWelcome and Introduction\n\nAntonino Tumeo and John Feo (Pacific Northwest National Laboratory (PNNL)); Michela Becch i (North Carolina State University); and Ana Lucia Verbanescu (University of Twente, Netherlands)\n---------------------\nIA^3 — Afternoon Break\n-- -------------------\nxBS-GNN: Accelerating Billion-Scale GNN Training on F PGA\n\nGraph Neural Networks (GNNs) have been used in a variety of challen ging applications. However, training GNN models is time-consuming as it in cur high volume of irregular data accessing due to its graph-structured in put data; such a challenge is further exacerbated in real-world applicatio ns as they ...\n\n\nYi-Chien Lin (University of Southern California (USC)) , Zhijie Xu (University of Michigan), and Viktor Prasanna (University of S outhern California (USC))\n---------------------\nAn Adaptive Asynchronous Approach for the Single-Source Shortest Paths Problem\n\nLarge-scale grap hs with billions and trillions of vertices and edges require efficient par allel algorithms for common graph problems, one of which is single-source shortest paths (SSSP). Bulk-synchronous parallel algorithms such as Delta- stepping experience large synchronization costs at the scale o...\n\n\nRit vik Rao, Kavitha Chandrasekar, and Laxmikant Kale (University of Illinois Urbana-Champaign)\n---------------------\nBatch Updates of Distributed Str eaming Graphs using Linear Algebra\n\nWe develop a distributed-memory para llel algorithm for performing batch updates on streaming graphs, where ver tices and edges are continuously added or removed. Our algorithm leverages distributed sparse matrices as the core data structures, utilizing equiva lent sparse matrix operations to execute g...\n\n\nElaheh Hassani, Md Tauf ique Hussain, and Ariful Azad (Indiana University)\n---------------------\ nPerformance Analysis of the NICAM Benchmark on MN-Core Processor\n\nLarge -scale Computational Fluid Dynamics (CFD) simulations are typical HPC appl ications that require both high memory bandwidth and large memory capacity . However, it is difficult to achieve high performance for such applicatio ns on modern high-performance processors due to their low memory bandwidt. ..\n\n\nHikaru Takayashiki and Natsuko Saito (Fixstars Corporation); Hirot o Imachi and Ryo Sakamoto (Preferred Networks); and Junichiro Makino (Pref erred Networks; Kobe University, Japan)\n---------------------\nNEO-DNND: Communication-Optimized Distributed Nearest Neighbor Graph Construction\n\ nGraph-based approximate nearest neighbor algorithms have shown high neigh bor structure representation quality.\nNN-Descent is a widely known graph- based approximate nearest neighbor (ANN) algorithm.\nHowever, graph-based approaches are memory- and time-consuming.\n\nTo address the drawbacks, we develop ...\n\n\nKeita Iwabuchi, Trevor Steil, Benjamin Priest, Roger Pea rce, and Geoffrey Sanders (Lawrence Livermore National Laboratory (LLNL))\ n\nTag: Graph Algorithms, Heterogeneous Computing, Programming Frameworks and System Software\n\nRegistration Category: Workshop Reg Pass\n\nSession Chairs: Michela Becchi (North Carolina State University); John Feo (Pacif ic Northwest National Laboratory (PNNL)); Antonino Tumeo (Pacific Northwes t National Laboratory (PNNL)); and Ana Lucia Varbanescu (University of Twe nte, Netherlands; University of Amsterdam, Netherlands) END:VEVENT END:VCALENDAR