Introduction to High Performance Computing for Life Scientists

We should all understand and follow the ARCHER2 Code of Conduct to ensure this course is conducted in the best teaching environment.

The course will be flexible to best meet the learning needs of the attendees.

Feedback is an essential part of our training to allow us to continue to improve and make sure the course is as useful as possible to attendees.

To be determined

Computer clusters enable users to run larger, more complex jobs in a reasonable time

HPC systems are state-of-the-art clusters composed of a large number of powerful computers linked by very fast interconnects.

ARCHER2’s login address is login.archer2.ac.uk.

The password policy for ARCHER2 is well documented.

There are a number of ways to transfer data to/from ARCHER2.

ARCHER2 consists of high performance login nodes, compute nodes, storage systems and interconnect.

There is a wide range of software available on ARCHER2.

The system is based on standard Linux with command line access.

Software is available through modules.

ARCHER2 uses the Slurm scheduler.

srun is used to launch parallel executables in batch job submission scripts.

There are a number of different partitions (queues) available.

To run an application in parallel we should understand how it uses threads and/or MPI and how the scheduler and parallel application launcher interact

Looking at parallel performance can tell us the most efficient way to run software to get a fast answer

As the number of cores are increased, performance rarely scales linearly.

Scalibility changes as the size and complexity of the system is changed.

Scalibility can be meaningless if the work load is not distributed evenly among processors.

Looking at parallel efficiency can help you make best use of computational resources

Most HPC systems run UNIX or Linux as an OS.

Each application is a seperate process; the OS will schedule any process.

Threads are sub-processes that compose a process.

To be determined

Larger systems scale better to large core-/node-counts than smaller systems.

To be determined

Hybrid MPI with OpenMP does affect performance.

When running hybrid jobs, placement across NUMA regions is important.

It can be benefical to the performance to use MPI+OpenMP

It can be benefical to the performance to use MPI+OpenMP

You can find more about the state of HPC in the UK here

You can learn more about HPC facilities across Europe on the PRACE and EuroHPC websites