The Best-Paper of IEEE/SBC SBAC PAD 2024 is:
"A new level-set analysis and sparse storage format for the SPTRSV in GPUs"
by Manuel Freire (Facultad de Ingeniería, UDELAR, Uruguay), Ernesto Dufrecho (Universidad de La República, Uruguay), and Pablo Ezzatti (Universidad de La República, Uruguay)
Chair: TBD
08:30 - 09:30
Keynote: Computer Architecture Research, the Road to Meaningful Contributions
Yale Patt (The University of Texas at Austin)
After almost 60 years in the trenches performing microarchitecture research, I have some thoughts about what it takes to move the needle. Most importantly, one must understand the hardware details that the microarchitecture is built on, and the details of the resulting microarchitecture. When it comes to bottom-up vs top-down, don't be a naive top-up. If you come up with a feature you feel the need to add, choose the microarchitecture, not the ISA. If the feature turns out to be a bad idea and you put it in the ISA, you will have to suffer with it for a very long time. Recognize the value of heterogeneity and embrace it enthusiastically. Take to heart the MIT white paper that there is plenty of room at the top and my response that there is still plenty of room at the bottom. Understand that there are some obstacles to performance that are caused by multiple problems and it is not until you nail the last problem that you will reap the benefit. Finally, always be willing to celebrate the good efforts of others; you are not the only smart person in the room. In this talk, I will try to expand on many of the points mentioned above.
09:30 - 10:00
Break
Wednesday - 10 am
Chair: Katarzyna Swirydowicz
10:00 - 10:25
TangramFP: Energy-Efficient, Bit-Parallel, Multiply-Accumulate for Deep Neural Networks *
Yuan Yao (Uppsala University, Sweden), Xiaoyue Chen (Uppsala University, Sweden), Hannah Atmer (Uppsala University, Sweden), Stefanos Kaxiras Uppsala University, Sweden)
10:25 - 10:50
DLDM: Locality-aware Data Management for accelerating Deep Learning Training
FNU Hariharan Devarajan (LLNL), Ian Lumsden (University of Tennessee, Knoxville), Chen Wang, Konstantia Georgouli, Jae-Seung Yeom, Michela Taufer
10:50 - 11:15
Janus: A Simple and Efficient Speculative Defense using Reinforcement Learning
Pavlos Aimoniotis (Uppsala University, Sweden), Stefanos Kaxiras (Uppsala University, Sweden)
11:15 - 11:40
Exploiting Processor Heterogeneity to Improve Throughput and Reduce Latency for Deep Neural Network Inference
Olivier Beaumont, Jean-François DAVID (Inria), Lionel Eyraud-Dubois, Samuel Thibault (Inria Bordeaux Sud-Ouest)
11:40 - 12:05
DeepWave: A Software Stack for Parallelizing Deep Learning Models Used in Geophysics
Allan Pinto (CNPEM, Brazil), Gustavo Leite (UNICAMP, Brazil), Marcio Pereira, Hervé Yviquel (UNICAMP, Brazil), Sandro Rigo (UNICAMP, Brazil), Guido Araujo (State University of Campinas, Brazil)
12:05 - 01:30
Lunch
Chair: TBD
01:30 - 02:30
Keynote: Democratizing AI: The Role of Open-Source Stacks
Priyanka Sharma (Fujitsu Research of India)
The present generation is witnessing the fastest industrial revolutions of all times and AI is its biggest enabler. With AI becoming pervasive in almost every domain of human existence, accessibility remains a challenge. With a mission to democratize AI, we need to work towards fostering open collaboration with the global community to develop technology that are human-centred and facilitate sustainable digital transformation. And while this requires an open mindset for learning and contributing as a professional, it also requires contributions towards developing an open-source AI software stack. This is important to driving innovation, democratizing access, fostering transparency, and ensuring ethical development of AI technologies. It empowers a global community of developers to collaborate and build upon each other's work, ultimately accelerating progress in the field.
Fujitsu’s 2nm Arm-based CPU, FUJITSU-MONAKA is focused on addressing the wide range of usage in the datacenters including the growing demand for AI and HPC workloads, while prioritizing sustainability and carbon neutrality.
Fujitsu’s unique microarchitecture is pivotal for CPU performance and power efficiency. Fugaku, the world’s most efficient supercomputer, is a testament to our technology. FUJITSU-MONAKA is poised to drive the next-gen AI application ecosystem with energy-efficient computing.
Our R&D efforts concentrate on enhancing various open-source AI stack and speeding up deep learning processes with various processor level optimizations.
Wednesday - 2:30 pm
Chair: Levy Scott
02:30 - 02:55
A new level-set analysis and sparse storage format for the SPTRSV in GPUs *
Manuel Freire (Facultad de Ingeniería, UDELAR, Uruguay), Ernesto Dufrecho (Universidad de La República, Uruguay), Pablo Ezzatti (Universidad de La República, Uruguay)
02:55 - 03:20
A Performance and Energy Study of GPU-Resident Preconditioners for Conjugate Gradient Solvers: In the Context of Existing and Novel Approaches *
Katarzyna Swirydowicz, Jesun Firoz (Pacific Northwest National Laboratory), Joseph Manzano, Stephen Thomas, Mahantesh Halappanavar, Kevin Barker
03:20 - 03:40
Break
Wednesday - 3:40 pm
Chair: Sandro Rigo
03:40 - 04:05
TailorFS: An Adaptive file system to support dynamic I/O requirements of HPC workloads
FNU Hariharan Devarajan (LLNL), Kathryn Mohror (Lawrence Livermore National Laboratory)
04:05 - 04:30
High-Quality I/O Bandwidth Prediction with Minimal Data via Transfer Learning Workflow
Dmytro Povaliaiev (RWTH Aachen University), Radita Liem (RWTH Aachen University), Julian Kunkel (GWDG), Jay Lofstead, Philip Carns
04:30 - 04:55
To Derive or Not to Derive: I/O Libraries Take Charge of Derived Quantities Computation
Ana Gainaru (Oak Ridge National Laboratory), Greg Eisenhauer, Norbert Podhorszki, Liz Dulac, Qian Gong (Oak Ridge National Laboratory), Anthony Kougkas, Jay Lofstead, Xian-He Sun (Illinois Institute of Technology, USA), Scott Klasky
04:55 - 05:20
Characterizing the Impact of Job Execution on the Occurrence of Memory Failures on a Petascale HPC System
Scott Levy (Sandia National Laboratories), Josh Hemmert, Kurt Ferreira, Kevin Pedretti
Chair: Guido Araujo
08:30 - 09:30
Keynote: Enhancing computer security with hardware-level malware detection
Jean Luc Gaudiot (University of California - Irvine, USA)
In the past decades, computer design has prioritized performance, cost reduction, and energy efficiency over security. Meanwhile, malicious attacks have surged with the ever-increasing number of Internet-connected devices. Traditional antivirus software struggles to combat these attacks, particularly those exploiting hardware vulnerabilities. We introduce an additional layer of malware detection at the hardware level, monitoring semantic and sub-semantic behaviors to enhance system security. We present a real-time malware detection system monitoring microarchitectural features to detect anomalies indicative of attacks like Rowhammer and Spectre. Our experiments demonstrate scalability and promising detection accuracy. Future research aims to extend detection to GPU and other hardware vulnerabilities, emphasizing proactive, multi-layered defense mechanisms to counter evolving malware threats.
09:30 - 10:00
Break
Thursday - 10 am
Chair: Stefanos Kaxiras
10:00 - 10:25
IDS-DEEP: a strategy for selecting the best IDS for Drones with heterogeneous EmbEdded Platforms *
Louis Morge-Rollet (ENSTAB), Camélia Slimani (ENSTAB), Laurent Lemarchand (University of Western Brittany), David Espes (Universite De Bretagne Occidentale, Brest), Frédéric Le Roy, Jalil Boukhobza (ENSTA)
10:25 - 10:50
Towards Performance Portability of an Oil and Gas Application on Heterogeneous Architectures
Arthur Francisco Lorenzon (UFRGS), Alexandre Sardinha (Petróleo Brasileiro S.A.), Bronson Messer, Philippe Olivier Alexandre Navaux (UFRGS)
10:50 - 11:15
Optimal Time and Energy-Aware Client Selection Algorithms for Federated Learning on Heterogeneous Resources
Alan Lira Nunes (UFF, Brazil), Cristina Boeres (Instituto de Computacao, UFF, Brazil), Lucia M. A. Drummond (UFF, Brazil), Laércio Lima Pilla (CNRS, France)
11:15 - 11:40
Disaggregated Memory with SmartNIC Offloading: a Case Study on Graph Processing
Jacob Wahlgren (KTH Royal Institute of Technology), Gabin Schieffer (KTH Royal Institute of Technology), Maya Gokhale, Roger Pearce, Ivy Peng (KTH Royal Institute of Technology)
11:40 - 12:05
Analyzing HPC Monitoring Data With a View Towards Efficient Resource Utilization
Samuel Maloney (FZJ), Estela Suarez (FZJ), Norbert Eicker (FZJ), Filipe Guimarães (FZJ), Wolfgang Frings (FZJ)
12:05 - 01:30
Lunch
01:30 - 02:30
Panel/Group Discussion: Sustainable HPC
Thursday - 2:30 pm
Chair: Philippe Navaux
02:30 - 02:55
DeVAS: Decoupled Virtual Address Spaces
Mirco Mannino (University of Siena, Italy), Biagio Peccerillo (University of Siena, Italy, Italy), Andrea Mondelli (Huawei, People's Republic of China), Sandro Bartolini (University of Siena, Italy)
02:55 - 03:20
DRAM errors and Cosmic Rays: Space Invaders or Science Fiction?
Isaac Boixaderas (Barcelona Supercomputing Center), Jorge Amaya, Sergi Moré, Javier Bartolome, David Vicente, Osman Unsal (Barcelona Supercomputing Center), Dimitris Gizopoulos (University of Athens), Paul M. Carpenter (Barcelona Supercomputing Center), Petar Radojkovic (Barcelona Supercomputing Center), Eduard Ayguade (Barcelona Supercomputing Center)
03:20 - 03:40
Break
Thursday - 3:40 pm
Chair: FNU Hariharan Devarajan
03:40 - 04:05
Memory Sandbox: A Versatile Tool for Analyzing and Optimizing HBM Performance in FPGA
Elias Perdomo (Barcelona Supercomputing Center), Xavier Martorell (Universitat Politècnica de Catalunya), Teresa Cervero (BSC-CNS), Behzad Salami (Barcelona Supercomputing Center)
04:05 - 04:30
S-Clflush: Securing Against Flush-based Cache Timing Side-Channel Attacks
Tejeshwar Bhagatsing Thorawade (Indian institute of Technology Bombay, India), Prajakta Yeola (Indian institute of Technology Bombay, India), Varun Venkitaraman (Indian Institute of Technology - Bombay, India), Virendra Singh
04:30 - 04:55
To Protect or Not To Protect: Probability-Aware Selective Protection for Sparse Iterative Solvers
Daniel Johnson, Hongyang Sun (University of Kansas, United States of America), Joshua Booth, Padma Raghavan
Chair: Alfredo Goldman
08:30 - 09:30
Keynote: Insights and Challenges for the Simulation of Parallel and Distributed Computing Systems and Applications: A SimGrid Perspective
Henri Casanova (University of Hawaii at Manoa)
The simulation of parallel and distributed computing systems has been actively pursued for several decades. The goal has been to develop methods and software to allow researchers in the field to perform perfectly controllable, observable and reproducible experiments for arbitary platform and application scenarios. The two key concerns with simulation are accuracy and scalability, and many approaches have been proposed and implemented to address these concerns in different ways. In this presentation we will review key historical developments, attempt to give a synthetic view of the state of the art, and highlight current challenges and possible solutions. Throughout the presentation, we will draw on key research and development milestones from the SimGrid project's extensive 20+ year history to offer a concrete and practical perspective.
09:30 - 09:50
Break
09:50 - 12:00
Session I - Tools for HPC Monitoring
10:00 - 10:30
Spinner: Enhancing HPC Experimentation with a Streamlined Parameter Sweep Tool
Rodrigo Ceccato de Freitas, Jhonatan Cléto, Gustavo Leite, Sandro Rigo, Jose M. Monsalve Diaz and Hervé Yviquel
Universidade Estadual de Campinas (UNICAMP - Brazil)
10:30 - 11:00
A Flexible Operational Framework for Energy Profiling of Programs
Roblex Nana Tchakoute, Claude Tadonki, Petr Dokladal and Youssef Mesri
Mines Paris - PSL (CRI/CMM/CEMEF - France)
Session II – Shared Memory Parallelism
11:00 - 11:30
An Instruction-Set Extension to Support Approximate Multicore Processors
Daniela Catelan, Felipe Sovernigo, Liana Duenha and Ricardo Santos
Federal University of Mato Grosso do Sul (UFMS - Brazil)
11:30 - 12:00
Experimental Study of Power Consumption of Basic Parallel Programs
Roblex Nana Tchakoute and Claude Tadonki
Mines Paris - PSL (CRI - France)
09:50 - 12:00
09:50
Opening
10:00
"User-Level Network Programmability: A Scalability Study for Data Center Infrastructure."
Patric Izolan, Ivan Lopes Junior, Ester Oribes, Braulio Souza, Angelo Crestani, Paulo Souza, Marcelo Luizelli, Fabio Rossi
10:20
"The Tracer Files: Crackling the Case of Performance Impact in Tracing Linux File I/O for I/O-intensive Applications."
Rodrigo Pereira Do Nascimento, Alfredo Goldman Vel Lejbman
10:40
"Leveraging Cloud Computing for Stock Market Forecasting with Reinforcement Learning."
Thiago Araújo, Arthur Lorenzon and Phillippe Navaux
11:00
"A Systematic Literature Review of I/O Optimization in HPC and Cloud Computing Environments."
Manuel Alexander Garcia Napa and Arthur Lorenzon
11:20
"Comparing Modern Differential Evolution Variants in Multiobjective Cloud Scheduling."
Pavel Kromer and Vojtech Uher
11:40
Final Considerations WCC
Papers: 15min + 5min Q&A
Best-Paper Candidate: the papers that have a * at the end of the title are best-paper candidates.