Overview

Teaching: 30 min
Exercises: 0 min

Questions

• What is High-Performance Computing (HPC)?

• What are the different options for HPC in MATLAB?

Objectives

• Understand what High-Performance Computing is.

• A brief overview different types of HPC techniques available to MATLAB users.

MATLAB

MATLAB is a multi-paradigm numerical computing platform for scientific computing, and is also a high-level programming language.

MATLAB offers support for linear algebra (Matrices and Vectors), function evaluations, plotting and visualisation, and symbolic computations for mathematical expressions.

Unlike C, C++ and Fortran, MATLAB is an interpreted language similar to Python. This means that programs written in MATLAB can be executed directly without the need for compiling them. Therefore, MATLAB is relatively easy to learn and use compared to compiled languages C, C++ and Fortran. However, the price one has to pay for this functionality is the slowness of computations. MATLAB is at least an order of magnitude slower than the programs written in C, C++ and Fortran. Nevertheless, MATLAB supports linking of C/C++ and Fortran codes with MATLAB code, if one is interested in replacing computing-intensive routines with C++ or Fortran programs.

Scope of this lesson

The main objective of this lesson is to learn how we can harness the computing power in High-Performance Computers using MATLAB. In this workshop we will focus on the techniques that help run MATLAB programs efficiently and effectively using command line interface and High-Performance Computing (HPC) resources. This material is aimed at problems that are computationally so demanding that each simulation runs for days on personal computers.

• While High-Performance Computing is considerably more involved in languages such as C, C++, Fortran and Python, it is relatively easier in MATLAB.
• MATLAB offers built-in support for High-Performance Computing in the form of parallelised loops, parallelised functions and the spmd construct within the MATLAB Parallel Computing Toolbox (PCT).
• In this course, we will focus on
  • Vectorization
  • Profiling
  • Parallel Computing Toolbox
    • parfor
    • spmd
    • mpiprofile
  • Running MATLAB from command line
  • Running MATLAB on Sunbird

What is High-Performance Computing (HPC)?

In lay terms, High-Performance Computing can be described as the process of performing more calculations in a given amount of time by utilising more computer hardware.

According to InsideHPC:

“High Performance Computing most generally refers to the practice of aggregating computing power in a way that delivers much higher performance than one could get out of a typical desktop computer or workstation in order to solve large problems in science, engineering, or business.”

High Performance Computing is carried out on a supercomputer, also called a cluster. A cluster is simply a group of computers with the same components (RAM, disk, processors/cores, and networking cards) as those in your desktop or laptop, but more powerful, and are networked with high-speed interconnect that can be accessed (indirectly) through software, the scheduler, that manages simultaneous execution of jobs by single or multiple users.

The following figure depicts a typical supercomputing cluster along with some of its important components. The user accesses the supercomputer by logging into the login node and by submitting jobs (or tasks) using jobscripts which will be managed by job schedulers such as Slurm Workload Manager, or PBS. The jobs are executed on the compute nodes. On Sunbird, we use the SLURM workload manager.

Nodes and Cores

Each individual computer in a cluster is commonly referred to as a “node”. Inside each node will be several processor chips that do the actual computation. A typical node in a cluster will have anything from 8 to 40 cores in total, often across several physical processor chips.

Reasons for using High-Performance Computing

• Your code takes a lot of time, for example, a couple of days, to run on your personal laptop or the desktop PC in your lab.
• Even though your code is fast enough you don’t have enough memory on your local machine to fit all of your data.
• Your code runs fast enough but you want to run 100s or 1000s of small jobs by varying some input parameters.
• You want to use a software library or proprietary software that you cannot install on your local machine for some reason.

MATLAB job scheduler

MATLAB has its own job scheduler. However, we do not recommend it on Supercomputing Wales clusters. We use SLURM.

MATLAB Computing speed

It is important to keep in mind the slowness of MATLAB codes. MATLAB codes are 10-100 times slower than the optimised C, C++ and Fortran codes. Since MATLAB is significantly slower than the code written in C, C++ and Fortran, parallelisation in MATLAB will also be considerably slower than what we can achieve in C, C++ and Fortran.

Key Points

• High-Performance Computing is all about expediting the computations.

• MATLAB offers several in-built functions to exploit the various HPC architectures, for example: multi-threads, multi-cores and GPUs, available today.

• MATLAB Parallel Computing Toolbox requires significantly less effort when compared with parallelisation in C, C++ and Fortran. But it comes at the cost of lower computing speeds.