Skip to content

Latest commit

 

History

History
86 lines (62 loc) · 2.3 KB

lesson-outline.md

File metadata and controls

86 lines (62 loc) · 2.3 KB
Raw
layout title
page
HPC Carpentry Lesson Outline

Lesson outline and todo list

This is the tentative list of tasks and topics for each lesson. Lesson writers are indicated with first/last initials (e.g. AR). Feel free to edit the topics for your section.

1. UNIX fundamentals - AR

  • SSH to a cluster

  • Bash fundamentals (cd, ls, ..., aliases, functions, ~/.bashrc)

  • Transferring files (scp? sftp? Maybe only one?)

  • Overview of HPC resources

    • What is a cloud?
    • What is a cluster? Different cluster types
    • Overview of services available (Compute Canada, Amazon EC2, etc.)

2. Submitting / running jobs - JS

  • Scheduler - lesson will cover SLURM (which can also run PBS scripts/commands natively)

    • Submitting jobs
    • Checking status of jobs
    • Deleting jobs
    • Job size consequences
    • GUI vs. batch programs (X-forwarding, SSH tunnels?)

  • Using software and environment modules

  • Playing friendly in the cluster

    • Understanding resource utilization
    • Profiling code - time, size, etc.
    • Getting system stats
    • Consequences of going over

3. Language refresher / introduction (Python - JB, Chapel - JZ+AR)

  • Programming language concepts

    • Compiled vs. interpreted languages

    • How does a program work?

    • Quick intro of programming language of choice

      • Major features + pros/cons
      • What is it good at?

  • Actual language overview

    • Basic syntax (arithmetic, variables, etc.)
    • Basic data structures (lists, arrays, etc.)
    • Defining functions
    • Conditional expressions
    • For-loops
    • Reading/writing data

Some side notes: perhaps a quick refresh of key concepts right before use in parallel section, certain concepts could get mixed in right before they're needed by the parallel lesson.

4. Intro to parallel programming (Python - JB, Chapel - JZ+AR)

  • Pipelining / automatic job submission / serial farming

  • Shared memory programming

  • Distributed memory programming

  • Overview of good parallel design

    • Dependencies within own code
    • Race conditions

  • Typical problems and bottlenecks

    • running in parallel (parallel scaling)
    • parallel I/O (don't write a 1GB file from one processor if data is already distributed, etc.)
    • Storage limitations (millions of files, compression, text vs. binary storage)
    • Filesystem choice (home, scratch, tmp, etc.)

Good luck!