- This event has passed.
aiXcelerate 2021
Monday, 6. December 2021 - Thursday, 9. December 2021
Topics: I/O & ML
Description
The annual workshop aiXcelerate at RWTH Aachen University (RWTH@NHR4CES) is a tuning workshop for HPC users. It comprises lectures that are open for everyone, and hands-on parts where registered groups apply the learnt concepts to their own codes. This year, aiXcelerate covers performance tuning for I/O-intensive workloads and machine-learning applications. It will take place online from December 6th to 9th, 2021.
The I/O part of the workshop examines the impact of unoptimized I/O on the overall application performance. The content covers I/O strategies, best practices, analysis tools, and hardware options from RWTH that can improve I/O performance. The ML part of the workshop addresses the increasing demand of computations tailored for GPUs. The content specifically focuses on how to detect whether the resources are efficiently utilized and in which situations it might suffice running workloads on the CPU partition to reduce the overall turnaround time.
Bring Your Own (BYO) Code – Please apply!
Since aiXcelerate is a tuning working, its purpose is to investigate your own code together with experts from the IT Center. If you are interested in a code analysis with respect to I/O or machine learning, please apply (informally) during the respective registration process: Let us know that you want to BYO code (and why) and we will contact you later with all details. We reserve the right to decline certain applications for BYO code.
Prerequiste for BYO: This offer targets at current and prospective users of the RWTH infrastructure. To this end, you should already have access to RWTH’s compute cluster CLAIX. Your code should run on CLAIX.
Organization
- There is no seminar fee.
- Presentations will be given in English. Slides will be available during or after the event.
- Please register separately for each part (I/O and ML) that you would like to attend.
- This is an online event. Detailled participation information (e.g., Zoom link) will be sent to registered participants shortly before the event.
Part I: I/O
Registration
Registration link for lectures and/or BYO code (December 6th – 7th, 2021): Registration is closed
Please select the corresponding check box if you are intersted in bringing your own code. If applicable, also tell us why you want to BYO (in the “comments” section).
Registration closing date: extended until Friday, November 26th, 2021
Skills
Course level: beginner to intermediate
Target audience
- HPC users
- HPC developers
Prerequisites
- Basic knowledge of Shell programming
- Basic knowledge of working in a Linux environment
Gained skills
- Choose the right filesystem for my application runs
- Understand the impact of I/O on my application’s runtime
- Execute basic I/O performance measurements with Darshan & Score-P/Vampir
- Basic knowledge of parallel I/O strategies and libraries available in HPC
Agenda
Monday, December 6th, 2021: I/O – Day 1
Time |
Topic |
Speaker
|
---|---|---|
09:00 – 10:00 | Does I/O matter to me? | Philipp Martin (RWTH) |
10:00 – 10:15 | Break | |
10:15 – 11:15 | Overview of I/O Technologies on CLAIX | Philipp Martin (RWTH) |
11:15 – 11:30 | Break | |
11:30 – 12:30 | Using Darshan for I/O Analysis | Radita Liem (RWTH) |
12:30 – 14:00 | Lunch Break | |
14:00 – 15:00 | Using Score-P & Vampir for I/O Analysis | Marc-André Hermanns (RWTH) |
15:00 – 15:15 | Break | |
15:15 – 16:45 | BYO: Preparation of benchmarks and job submissions for user codes |
Tuesday, December 7th, 2021: I/O – Day 2
Time |
Topic |
Speaker
|
---|---|---|
09:00 – 10:00 | I/O Libraries: Overview and MPI-IO | Marc-André Hermanns (RWTH) |
10:00 – 10:15 | Break | |
10:15 – 11:00 | I/O Libraries: HDF5 | Sebastian Lührs (Forschungszentrum Jülich) |
11:00 – 11:15 | Break | |
11:15 – 12:00 | I/O Patterns Best Practice | Radita Liem (RWTH) |
12:00 – 13:30 | Lunch Break | |
13:30 – 15:00 | BYO: Review benchmark results | |
15:00 – 15:15 | Break | |
15:15 – 16:00 | BYO: Lightning talks about take-aways |
Course Material
- 2021-aiXcelerate-Organization
- 2021-aiXcelerate-IO-01-Does-IO-Matter-To-Me
- 2021-aiXcelerate-IO-02-Overview-of-IO-Technologies-on-CLAIX
- 2021-aiXcelerate-IO-03-Darshan-Tutorial
- 2021-aiXcelerate-IO-04-Score-P-Cube-Vampir
- 2021-aiXcelerate-IO-05-IO-Schemes
- 2021-aiXcelerate-IO-06-MPI-IO-presentation
- 2021-aiXcelerate-IO-07-HDF5
- 2021-aiXcelerate-IO-07-HDF5-example
- 2021-aiXcelerate-IO-08-IO-Best-Practices
Part II: Machine Learning (ML)
Registration
Registration link for lectures and/or BYO code (December 8th – 9th, 2021): Registration is closed
Please select the corresponding check box if you are intersted in bringing your own code. If applicable, also tell us why you want to BYO (in the “comments” section).
Registration closing date: extended until Friday, November 26th, 2021
Skills
Course level: beginner to intermediate
Target audience
- HPC users
- HPC developers
Prerequisites
- Basic knowledge of Shell and Python programming
- Basic knowledge of working in a Linux environment
Gained skills
- Understanding the challenge of turnaround times on CLAIX due to overbooked GPU resources
- Select appropiate cluster resources and problem sizes for development, testing and production phases
- Determine basic resource utilization of my ML/DL application
- Configure and run my desired ML/DL execution environment (e.g. by utilizing existing containers on CLAIX)
- Choose the right filesystem for my ML/DL application runs
Agenda
Wednesday, December 8th, 2021: ML – Day 1
Time |
Topic |
Speaker
|
---|---|---|
09:00 – 09:30 | Welcome & ML Focus | Jannis Klinkenberg (RWTH) |
09:30 – 10:00 | Infrastructure Overview and Future Plans | Christian Terboven (RWTH) |
10:00 – 10:15 | Break | |
10:15 – 11:00 | Vision: Development, Testing and Production Runs | Jannis Klinkenberg (RWTH) |
11:00 – 11:15 | Break | |
11:15 – 12:30 | Configuring and Running ML/DL workloads on CLAIX | Sven Hansen (RWTH), Jannis Klinkenberg (RWTH) |
12:30 – 14:00 | Lunch Break | |
14:00 – 14:30 | Identifying whether my workload makes efficient use of resources | Jannis Klinkenberg (RWTH) Philipp Martin (RWTH) |
14:30 – 14:40 | CLAIX Support for Distributed Machine Learning | Radita Liem (RWTH) |
14:40 – 15:00 | Case Study: Performance Analysis of Large Scale Distributed DL | Anara Kozhokanova (RWTH) |
15:00 – 15:15 | Break | |
15:15 – 17:00 | Hands-on: Running ML/DL Workloads on CLAIX |
Thursday, December 9th, 2021: ML – Day 2
Time |
Topic |
Speaker
|
---|---|---|
09:00 – 16:00 | BYO |
Course Material
- 2021-aiXcelerate-Organization
- 2021-aiXcelerate-ML-01-Agenda-and-Infrastructure-Overview
- 2021-aiXcelerate-ML-02-Motivation-and-Vision
- 2021-aiXcelerate-ML-03-Configuring-and-Running
- 2021-aiXcelerate-ML-04-Identifying-Resource-Utilization
- 2021-aiXcelerate-ML-04b-Distributed-DL-Support
- 2021-aiXcelerate-ML-04c-Case-Study
- 2021-aiXcelerate-ML-05-Hands-on
- 2021-aiXcelerate-hands-on-ML.tar
- 2021-aiXcelerate-ML-06-Further-Links-and-Literature