Explore multi-threading in Julia with PARTR in this 26-minute conference talk from JuliaCon 2019. Dive into the evolution of threading in Julia, covering past implementations, current features, and future developments. Learn through practical examples including parallel merge sort, Prime Sieve of Eratosthenes, and prefix scan, with performance comparisons between parallel and single-threaded implementations. Gain insights into PARTR's relationship with threading, scheduling dynamics, thread spawning limits, and alternative threading models. Address key questions about distributed computing, low-level details, composition with pmap, scheduler priorities, and BLAS native threads. Enhance your understanding of Julia's multi-threading capabilities and their practical applications in high-performance computing.
Overview
Syllabus
Welcome!.
Outline.
What's old?.
What's new?.
What's coming?.
Multithreading examples.
Examples - Parallel merge sort.
Parallel vs. Single-thread performance (Sort).
Examples - Prime Sieve of Eratosthenes.
Parallel vs. Single-thread performance (Prime Sieve).
Examples - Prefix scan.
Parallel vs. Single-thread performance (Prefix Scan).
Question - How does this relate to PARTR?.
Question - Is the schedule dynamic?.
Question - How many threads can be spawned?.
Question - Have you thought about alternative threading models?.
Question - Is there a reason to use Distributed instead of Threads?.
Question - Some low-level details.
Question - Does it compose with pmap?.
Question - Will the scheduler implement priorities?.
Question - What about BLAS native threads?.
Wrap-up.
Taught by
The Julia Programming Language
