Parallel R Workflows: Welcome!

---

Introduction

Dr Anna Krystallli

R-RSE

https://parallel-r.netlify.app

👋 Hello

me: Dr Anna Krystalli

• Research Software Engineering Consultant, r-rse

  • mastodon annakrystalli@fosstodon.org
  • twitter @annakrystalli
  • github @annakrystalli
  • email r.rse.eu[at]gmail.com

• Editor rOpenSci

• Founder & Core Team member ReproHack

Objectives

In this course we’ll explore:

• General concepts and strategies to parallelisation.

• R packages available for parallelising computation in R.

• The futureverse collection of packages.

• Deploying Parallel workflows on the Iridis cluster

Background

Computation

Transistor icons created by surang - Flaticon

Computers represent info using binary code in the form of digital 1s and 0s inside the central processing unit (CPU) and RAM. These digital numbers are electrical signals that are either on or off inside the CPU or RAM.

Each transistor is a switch, that is, 0 when turned off and 1 when turned on. The more transistors, the more switches.

Transistors are the basic building blocks that regulate the operation of computers, mobile phones, and all other modern electronic circuits and is the basic unit of the CPU.

The more transistors you have, the more computation you can perform.

Moore’s law

Moore’s Law, which was stated by Gordon Moore, one of the co-founders of Intel, in 1975 stated that

if the price is unchanged, the number of transistors on a microchip (CPU, GPU, RAM) will double every 18-24 months, meaning performance will double as well.

This has indeed played out as we can see in this graph as well as the next one.

Yet…

we’ve hit clock speed stagnation

50 Years of Processor Trends. Distributed by Karl Rupp under a CC-BY 4.0 License

Yet…we’ve hit clock speed stagnation

Clock speed: measured in MHz represents how many cycles a CPU can execute per second, Clock speed stagnation: refers to the slowdown or halting of increases in the clock speeds of computer processors, which occurred in mid-00s after decades of steady improvement.

• that’s beacuse As clock speeds increase, so does power consumption and heat generation and we also start running into quantum and material constraints. So due to heat dissipation challenges, quantum effects, and manufacturing difficulties We have effectively approached practical limit for the number of transistors that can be fitted into a single core.

To maintain performance growth, CPU manufacturers shifted their focus to improving multi-core architectures, allowing multiple processes to run simultaneously.

This shift means modern performance gains rely on optimized software capable of leveraging parallel processing rather than simply relying on raw clock speed increases. so while in the past devs could sit back and wait for hardware to catch up, now they need to write their code in a way that can take advantage of parallel processing. Hence this course!

About computer hardware

CPU (Processing)

RAM (memory)

Hard Disks, Networks (I/O)

CPU

• The central processing unit (CPU), or the processor, is the brains of a computer. The CPU is responsible for performing numerical calculations.

• The faster the processor, the faster R will run.

• If your analysis requires a lot of computing power we say it is CPU-bound.

RAM

• Random access memory (RAM) is a type of computer memory that can be accessed randomly: any byte of memory can be accessed without touching the preceding bytes.

• The amount of RAM R has access to is incredibly important. Since R loads objects into RAM, the amount of RAM you have available can limit the size of data set you can analyse. we call workflows that put a lot of pressure on memory MEMORY BOUND

Even if the original data set is relatively small, your analysis can generate large objects

About R

R is an interpreted language

Compiled Languages

Converted directly into machine code that the processor can execute.

• Tend to be faster and more efficient to execute.

• Need a “build” step which builds for system they are run on

• Examples: C, C++, Erlang, Haskell, Rust, and Go

Interpreted Languages

Code interpreted line by line during run time.

• significantly slower although just-in-time compilation is closing that gap.

• much more compact and easy to write

• Examples: R, Ruby, Python, and JavaScript.

R performance

• R offers some excellent features: dynamic typing, lazy functional evaluation and object-orientation

• Side effect: operations are undertaken in single-threaded mode, i.e. sequentially

• Many routines in R are written in compiled languages like C & Fortran.

• R performance can be enhanced by linking to optimised Linear Algebra Libraries.

• Many packages wrap more performant C, Fortran, C++ code.

• R offers many ways to parallelise computations.

R as a language and environment is reasonably well established and understood. A combination of dynamic typing, lazy functional evaluation and object-orientation (in several flavors) makes for a somewhat unique combination.

One side-effect of this design is that core operations are undertaken in single-threaded mode, or, in other words, sequentially.

About this course

• I normally like to live code…BUT!

• There’s a lot of materials to get through so I will be copying & pasting from the materials alot

• Have the materials handy to follow along

• Please stop me for questions or to share your own experiences

• Lunch around 1pm

Let’s go!