Fabian Renn-Giles & Dave Rowland - Real-time 101 - part I: Investigating the real-time problem space

Thank you to our VIP patrons:
Ahmet Levent Tasel
Art and Logic
Auxy
Elk Audio
Felipe Tonello
Glenn Kasten
Inphonik
Jerry Chan
Larry Mickie
Matt Gilg
Overloud
Simon Holt
Sound Radix
Steve Flower
Tatsuya Shiozawa
Thomas Chatfield
UJAM
Thank you to our All-access patrons
Adam Gazda
Alain Martin
Alex
Alexander Gustafson
Andy Coenen
Anthony Lopez
Athan Billias
Bence Kovacs
Bome Software
Celso Gomes
Charles
Daniel Protheroe
Daniel Walz
Danjeli Schembri
Dave Foster
David Groff
Devendra
Dominic Curtis
Don Turner
Eyal Amir
Fabian Renn
Gen Hori
Graeme MacDonald
Greg Cerveny
Greg Olin
Griffin Moe
Halil Kleinmann
Jerome Noel
Jian Yu
Jim Knowler
Jim Lochowitz
John Carlson
Julius O Smith
Kevin Vanwulpen
Lee Huddleston
Marcus Ficner
Oli Larkin
Oliver Greschke
Paul Zielie
Ray Cernis
Rory Walsh
Taro
Tim Mayberry
Vincent Meyer
廣太郎 佐々木
Description: “Real-time” is a term that gets used a lot, but what does it really mean to different industries? What happens when our “real-time” system doesn’t perform in real-time? And how can we ensure that we don’t get in to this situation?
This talk aims to discuss what we mean by a real-time system, the practices that can be used to try and make sure it stays real-time and in particular how these can be subtly or accidentally abused increasing the risk of violating your real-time constraints.
We’ll take a detailed look at some of the considerations for real-time systems and the costs they involve such as system calls, allocations and priority inversion. Then the common tools in a programmer’s box such as mutexes, condition variables and atomics, how these interact with real-time threads and what costs they can incur.
This talk aims to ensure attendees of all experience levels leave with a solid understanding of the problems in the real-time domain and an overview of the tools commonly used to solve them.
Twitter: / audiodevcon
More info: juce.com/adc

[@FabianRenn]

I'd like to point out that I answered the last question incorrectly: most architectures will ensure that the cache is coherent between multiple cores, i.e. if a cache line is written to by one core, another core reading the same cache line will see the updated contents. I stated in the talk that it could potentially see an old value - this is wrong on most common architectures. Sorry for the confusion.

[@vapon]

20:40 Scott Meyers mentions that make_shared() actually creates control block and the instance in a single memory block

[@Parawaveaudio-com]

The flow chart is a really nice summary of all the things to consider when dealing with this kind of stuff.
Would be great to see some of the farbot stuff in a juce module, or even include in into juce_core.

[@claudiostano]

This lecture is great !! Thank you !!