High Performance Computing. ISC High Performance 2022 International Workshops: Hamburg, Germany, May 29 – June 2, 2022, Revised Selected Papers

✍ Scribed by Hartwig Anzt, Amanda Bienz, Piotr Łuszczek, Marc Baboulin

• Publisher: Springer
• Year: 2023
• Tongue: English
• Leaves: 398
• Series: Lecture Notes in Computer Science, 13387
• Category: Library

✦ Synopsis

This book constitutes the refereed conference proceedings of the workshops held at the 37th International ISC High Performance 2022 Conference, in Hamburg, Germany, in June 2, 2022.
The 27 full papers were included in this book were carefully reviewed and selected from 43 submissions. 
ISC High Performance 2022 presents the following workshops:
Compiler-assisted Correctness Checking and Performance Optimization for HPC
HPC on Heterogeneous Hardware (H3)

Malleability Techniques Applications in High Performance Computing

Fifth Workshop on Interactive High Performance Computing
3rd ISC HPC International Workshop on Monitoring & Operational Data Analytics
6th International Workshop on In Situ Visualization
17th Workshop on Virtualization in High Performance Cloud Computing

✦ Table of Contents

Preface
Organization
Contents
Compiler-Assisted Correctness Checking and Performance Optimization for HPC
Preface to the Third Workshop on Compiler-Assisted Correctness Checking and Performance Optimization for HPC (C3PO’22)
1 Introduction
2 Organization
2.1 Organizing Committee
2.2 Program Committee
3 Program
3.1 Invited Talk
3.2 Research Papers
Compiler-Assisted Instrumentation Selection for Large-Scale C++ Codes
1 Introduction
2 Related Work
3 Tailored Instrumentation for OpenFOAM
3.1 Design and Limitations of InstRO
4 The CaPI Instrumentation Toolchain
4.1 Instrumentation Workflow
4.2 Implementation
4.3 Score-P Integration
5 Evaluation on OpenFOAM
6 Usability and Validation Impediments
7 Discussion
8 Conclusion and Future Work
References
Lightweight Array Contraction by Trace-Based Polyhedral Analysis
1 Introduction
2 Background
2.1 Polyhedral Model
2.2 Array Contraction
3 Related Work
4 Our Approach
4.1 Overview of the Approach
4.2 Generating Input Parameter Instances
4.3 Inferring a Mapping on a Trace
4.4 Interpolation
5 Experimental Results
5.1 Experimental Setup
5.2 Results
6 Conclusion
References
Detecting Scale-Induced Overflow Bugs in Production HPC Codes
1 Introduction
2 Tracing Algorithm Extension
2.1 Fortran Support
3 Evaluation
4 Related Work
5 Conclusion
References
HPC on Heterogeneous Hardware (H3)
AI Benchmarking for Science: Efforts from the MLCommons Science Working Group
1 Introduction
2 MLCommons Science Working Group
2.1 About the Working Group
2.2 Science Benchmarking
2.3 Policies for Benchmarking
3 Benchmarks for the First Release
3.1 Cloud Masking (cloud-mask)
3.2 STEMDL (stemdl)
3.3 CANDLE-UNO (candle-uno)
3.4 Time Series Evolution Operator (tevelop)
4 Results from Initial Evaluations
4.1 Results for the cloud-mask Benchmark
4.2 Results for the stemdl Benchmark
4.3 Results for the candle-uno Benchmark
4.4 Results for the tevelop Benchmark
5 Conclusions
References
Performance Analysis of Matrix Multiplication for Deep Learning on the Edge
1 Introduction
2 Blocked Algorithms for GEMM
2.1 The Baseline Algorithm for GEMM
2.2 A Family of Algorithms for GEMM
3 A Performance Simulator for GEMM Algorithms
3.1 IoT Architecture Model
3.2 Validation
4 Performance Analysis
5 Discussion and Future Work
References
Strategies for Efficient Execution of Pipelined Conjugate Gradient Method on GPU Systems
1 Introduction
2 Related Work
3 Background
4 Methodology
4.1 Hybrid-PIPECG-1 Method
4.2 Hybrid-PIPECG-2 Method
4.3 Hybrid-PIPECG-3 Method
5 Experiments and Results
6 Conclusion and Future Work
References
A Multi-Level Platform-Independent GPU API for High-Level Programming Models
1 Introduction
2 Background
2.1 Chapel
2.2 Chapel's GPUIterator Module
3 Design
3.1 Motivation
3.2 MID-LOW-level API: Thin Wrappers for Raw GPU Routines
3.3 MID-level API: A Chapel Programmer Friendly GPU API
3.4 Supporting Asynchrony
4 Implementation
4.1 Library Implementation
4.2 The GPU Kernel Part by the User
5 Performance and Productivity Evaluations
5.1 Distributed Mini Applications
5.2 Real-world Example: Distributed Tree Search
6 Related Work
7 Conclusions
References
Precise Energy Consumption Measurements of Heterogeneous Artificial Intelligence Workloads
1 Introduction
2 Related Work
3 Experimental Evaluation
3.1 Workloads
3.2 Computation Environment
3.3 Measurement Setup
4 Results
5 Conclusion
References
Malleability Techniques Applications in High Performance Computing
Malleability Techniques and Applications in High-Performance Computing (HPCMALL 2022)
1 Motivation and Objectives
2 Topics
3 Contributions
4 Workshop organization
4.1 Workshop Chairs
4.2 Program Committee
5 Workshop background and perspective
6 Acknowledgements
Detecting Interference Between Applications and Improving the Scheduling Using Malleable Application Proxies
1 Introduction
2 Monitor Architecture
2.1 System Monitor
3 Building Synthetic Micro-benchmarks
3.1 Application for Improving Machine Learning Algorithms
3.2 Application for Application Interference Analysis
4 Evaluation
4.1 Proxy Accuracy
4.2 Prediction Algorithms Improvement
4.3 Interference Detection Using Malleable Proxies
5 Related Work
6 Conclusion
References
An Emulation Layer for Dynamic Resources with MPI Sessions*-10pt
1 Introduction
1.1 Motivation
1.2 Related Work
2 MPI Sessions
3 Dynamic Resources with MPI Sessions
3.1 MPI Sessions Advantages Compared to MPI_COMM_WORLD
3.2 Resource Changes with Process Sets
4 Interface Design
4.1 MPI Sessions Interface
4.2 MPI_Session_get_psets
4.3 Process Set Management Interface
4.4 Resource Change Management Interface
5 libmpidynres
5.1 libmpidynres as an Emulation Layer on Top of MPI
5.2 Emulated Process States
5.3 Resource Manager
6 Case Study
7 Conclusion
References
Exploiting OpenMP Malleability with Free Agent Threads and DLB
1 Introduction
2 Related Work
3 Implementation
3.1 The LLVM OpenMP Runtime
3.2 The Role-Shifting Threads
3.3 Integration with DLB
4 Evaluation
4.1 HPC Environment
4.2 ParMmg
4.3 Alya
5 Conclusions
References
QR Factorization Using Malleable BLAS on Multicore Processors
1 Introduction
2 Exploiting Parallelism in Task-Parallel Applications
3 QR Factorization with Look-Ahead
4 Performance Evaluation
5 Conclusions
References
IMSS: In-Memory Storage System for Data Intensive Applications
1 Introduction
2 Related Work
3 IMSS Architecture Design
3.1 Front-End Layer
3.2 Back-End Layer
3.3 IMSS Metadata Server
3.4 Data Distribution Policies
4 Deployment Strategies
5 Experimental Evaluation
5.1 Block Size Variation
5.2 Scalability
5.3 Metadata Overhead
6 Conclusions
References
On the Convergence of Malleability and the HPC PowerStack: Exploiting Dynamism in Over-Provisioned and Power-Constrained HPC Systems
1 Introduction
2 Technology Trend
2.1 Hardware Architecture
2.2 Malleability Support
2.3 Co-scheduling
2.4 Power-Aware HPC
3 Problem Statement
3.1 Job Classification
3.2 Malleable Jobs Under Power Constraint
4 Toward Convergence of Malleability and PowerStack
4.1 Strawman Architecture
4.2 Solution Overview and Requirements
4.3 Our Ongoing Efforts on Software Tool Integration
5 Conclusion
References
The Fifth Workshop on Interactive High Performance Computing
Interactive, Cloud-Native Workflows on HPC Using KNoC
1 Introduction
2 Related Work
3 Design
4 Implementation
4.1 The KNoC Virtual Kubelet Provider
4.2 The Door Executable
4.3 Integration with Argo Workflows
5 Evaluation
6 Conclusion
References
Workflows to Driving High-Performance Interactive Supercomputing for Urgent Decision Making
1 Introduction
2 Background
3 The Anatomy of Our Workflows
3.1 Marshalling and Control System Workflows
3.2 HPC Machine Side Workflows
4 Case-Study: Interactive Urgent Space Weather Ensembles
5 Conclusions
References
The 3rd ISC HPC International Workshop on Monitoring and Operational Data Analytics
The 3rd International Workshop on Monitoring and Data Analytics (MODA22)
1 Introduction
2 Workshop Organisation
2.1 Organising Committee
Workshop Chairs
Publicity Chairs
Program Committee
3 Conclusion
Data Center Facility Monitoring with Physics Aware Approach
1 Introduction
2 Data Collection Infrastructure
3 Data Center Anomaly Detection and Visualization
4 Physics-Informed Anomaly Monitoring
5 Conclusions
References
Rule-Based Thermal Anomaly Detection for Tier-0 HPC Systems
1 Introduction
1.1 Background
2 Related Work
3 Methodology
3.1 Dataset
3.2 Rule-Based Statistical Method (Flags)
3.3 Mathematical Definition of the Flags
4 Experimental Results
4.1 Severity Level of Anomaly (Flags)
4.2 Detailed Study of Real Physical Failure
4.3 Locations of Anomalies
5 Summary and Future Work
References
The 6th International Workshop on In Situ Visualization
The 6th International Workshop on In Situ Visualization (WOIV’21)
1 Background and Description
2 Workshop Summary
2.1 Keynote
2.2 Capstone
2.3 Papers
3 Organising Committee
3.1 Workshop Chairs
3.2 Workshop Co-organizers
3.3 Program Committee
In Situ Analysis and Visualization of Extreme-Scale Particle Simulations
1 Introduction
2 Related Works
3 ParaView Catalyst-based in Situ Visual Analysis Workflow
3.1 In Situ Catalyst Adapter Design
3.2 In Situ Particle Density Estimation for Effective Visualization of Data Features
4 Evaluation
5 Conclusions
References
Insite: A Pipeline Enabling In-Transit Visualization and Analysis for Neuronal Network Simulations
1 Introduction
2 Related Work
3 Method
4 Application
5 Performance
6 Conclusion and Future Work
References
The Need for Pervasive In Situ Analysis and Visualization (P-ISAV)
1 Introduction
2 Related Work
3 Our Position
3.1 Suggested Research Directions
4 Conclusion and Future Directions
References
Interactive Visualization of Large-Scale Oil and Gas Reservoir Simulation Models
1 Introduction
2 Visualization Workflow
3 Performance Testing
4 Discussion
5 Conclusions
References
Cinema Transfer: A Containerized Visualization Workflow
1 Introduction
2 Related Work
3 Workflow Description
4 System Configuration
4.1 Host Configuration
4.2 Message Passing Interface in Containers
4.3 Embedded MPI with Kubernetes
4.4 Hybrid MPI with Singularity
4.5 Tmpfs Filesystems
4.6 Transferring Cinema Databases
5 Evaluation
5.1 Embedded MPI with Kubernetes, SENSEI In Situ
5.2 Hybrid MPI with Singularity, SENSEI In Situ
5.3 Hybrid MPI with Singularity, SENSEI In Transit
5.4 Cinema_transfer Loop Time
6 Discussion
7 Conclusion
References
The 17th Workshop on Virtualization in High Performance Cloud Computing
Virtual Clusters: Isolated, Containerized HPC Environments in Kubernetes
1 Introduction
2 Related Work
3 Design Overview
4 Implementation
5 Evaluation
6 Conclusion
References
Analyzing Unikernel Support for HPC: Experimental Study of OpenMP
1 Introduction
2 A Background on Unikernels
3 Experimental Setup
4 Stability Evaluation
5 Performance Evaluation
6 Conclusion
References
On the Use of Linux Real-Time Features for RAN Packet Processing in Cloud Environments
1 Introduction
2 Scheduling the BaseBand Application
3 CPU Configuration
4 Experimental Results
5 Conclusions
References
eBPF-based Extensible Paravirtualization
1 Introduction
2 eBPF
3 Extensible Paravirtualization
4 Virtual to Physical CPUs Affinity
4.1 Implementation
4.2 Tests
5 Results
5.1 Virtual CPU Pinning
5.2 Virtual Hyper-thread Pinning
6 Confidential Computing
7 Conclusions and Future Work
References
Author Index