Practical Scientific Computing

A seminar providing the required essentials — and more — to leverage shared resources efficiently.

Authors

Dr. Jonas I. Liechti
Dr. Matteo Delucchi

Content

About

• About this Seminar
  • Syllabus

Part 1: Utilizing Shared Resources

• Challenges in Computational Projects
  • Introduction
  • Storage Limitations
  • Accessibility and Data Movement
  • Working with Large Datasets
  • Computational Constraints
  • Task Multiplicity & Parallelism
  • Transparency & Reproducibility
• Primer on Computer
  • Universal Turing Machine
  • Z3 Computer
  • The Stored-program Computer
  • The Transistor Computer
  • Integrated Circuits & MOSFET
  • Microprocessors
  • The scaling laws
  • Physical limits
  • The Power Wall
• From Megahertz to Teraflops
  • How kept computational power growing?
  • Challenges
  • Efficiency through complexity
• Concurrency
  • Definition
  • Issues & Challenges
  • Parallelism
  • Processes vs. Threads vs. Coroutines
  • Architectural Levels of Parallelism
• From Concurrency to Service
  • Multitenancy
  • Resource sharing
  • as a Service
• as a Service
  • Usage & Origins
  • Infrastructure as a Service
  • HPC as a Service
  • Storage as a Service
• Virtualization
  • Types
  • Virtual Machine (VM)
• The Cloud
  • Cloud Computing Schema
  • Cloud Features
  • Cloud Operating System
• HPC Clusters
  • HPC Cluster Schema
  • HPC Cluster Features
  • Slurm Cluster Architecture
• Distributed Storage
  • Types of Cloud Storage
• Terminology
  • Cluster
  • High Performance Computing (HPC)
  • The “Shared” Reality
  • Slurm
  • Cloud Operating System
  • Virtualization
  • Container
  • Dockerfile

Part 2: Research Software Engineering

• Computational Project
  • Definition
  • Good Scientific Practice
• 4 Pillars of SoC
  • The Four Pillars of SoC
  • Project Scaffolding
  • Template Repository
• Introduction to Version Control
  • Why Use Version Control?
  • Git Basics
  • Basic Git Workflow
  • Versioning
  • Further Resources and Topics
• Project Documentation
  • Why Document?
  • The README.md File
  • Project Title & Description
  • Installation & Usage & Methodology
  • Data & Licensing & Citation
  • Acknowledgments & Contributing & Contact
• Extended Documentation
  • Extended Documentation Requirements
  • Documentation Tools and Hosting
  • Building Documentation with Sphinx
• Managing Data
  • Linking Data to a Project
  • Versioning Data 1.0
  • Versioning Data 2.0
  • Accessing Object Storage
  • Managing Hugging Face Models
• Code Structure
  • Codebase Architecture
  • The src/ Layout
  • Testing and Documentation
  • Dependency Separation via pyproject.toml
  • Container Blueprints
• Environments 1.0
  • Python Environments
• Primer on Software Development
  • Why Software Development Principles?
  • Good Coding Practices
  • Orthogonal Code
  • The DRY Principle
  • Single Source of Truth (SSOT)
• Software Development Strategies
  • Rubber Ducking
  • Functions & Classes
  • Unit Testing
• Refactoring
  • Definition
• Notes on Python
• Efficiency in Python
  • Slow & Popular
  • Python’s Elephant in the Room
  • Gotcha 1: Parallelization
  • Gotcha 2: Context Switches
  • Profiling and Debugging
• Primer on Parallelism
• Primer On Parallelism
• Primer on Parallelism
  • Implementing Parallelism: Multi-Threading vs. Multi-Processing
  • Parallelism in Python: Capabilities and Constraints
• Environments 2.0
  • Virtualized Environments

Part 3: Recap; Containers, Clouds & Objects

• Recap: Last Week
  • Recap the Morning
  • Questions from Last Week?
  • Virtualization
  • The Cloud
  • The HPC Cluster
  • Distributed Storage
  • Recap the Afternoon
  • Project Root Structure
  • The 4 Pillars of Separation of Concerns
  • SoC: Drain Your Scripts!
  • Version Control
  • Software Design Principles
  • Python Gotcha: Nested Parallelism
• Recap: exoHelloWorld
  • Building & Running
  • What goes into the container?
  • What needs to be build inside?
  • How to prepare the runtime environment?
  • What should the container do?
  • Binds at Runtime
• Example .def #1
• Example .def #2
• Example .def #3:
• Apptainer Runtime Guide
• run vs exec vs shell
• Suggested container definition
• Containers for SoC projects
  • Environment Management
  • Concern Substitution
• CI/CD Build
• Fetching a Container at Runtime
  • Access and Authentication
  • Execution via ORAS
• The UZH Science Cloud
  • Creating a VM via the Web GUI
  • CLI Access via openstack
• Ansible Detour
  • Getting Started with Ansible
  • One Shot Config
• Swift Object Storage
  • Accessing Object Storage

Part 4: Cluster - Workflow & Profiling

• Recap: HPC Clusters
  • The HPC Cluster
  • Shared Parallel Filesystem
  • Features & Limitations
  • How to Interact
  • SSH Agent Forwarding
• Containers for HPC
  • The Principle of OS-Level Virtualization
• The HPC Cluster Environment
• System-Level Packages
• The “Invisible” Container
  • The Hidden Cost
  • The Definition File
  • Using the Environment
• The “Invisible” Container
  • The Hidden Cost
• HPC Clusters at UZH
  • At a Glance
  • The Science Cluster
  • Science Cluster Storage
  • The Supercomputer (Alps)
  • Accessing the Systems
• Login and Look Around
  • Helpful Commands
• Latency and Throughput
  • Connection Fabric
  • GPU Communication Overview
  • GPU-NVMe Latency Comparison
• Profiling
  • Tools
  • Runtime
  • 👷 Practical Part 👷
• Special Topics
  • Using Shared Memory (/dev/shm)

Outro

• Outro
  • Content
  • Instructors
  • About