Embark on a Star Trek-themed journey through queuing theory and application scaling in this 30-minute RailsConf 2022 talk. Explore the distinctions between parallelism and concurrency, and learn when to implement additional threads or processes. Dive into examples using Sidekiq and Puma with various concurrency settings in different IO scenarios. Discover key concepts such as Little's Law, the USE Method, vertical and horizontal scaling, and Amdahl's Law. Examine job queueing with Sidekiq, request queueing with Puma, and intelligent routing techniques. Gain insights into balancing customer satisfaction and system performance while optimizing your Ruby applications for scalability and efficiency.
Overview
Syllabus
Intro
Transporters Capacity Planning
Basic Terminology Queueing Things
Little's Law
USE Method - Intel Engineer Brendan Gregg
Saturation Causes Service Delays
Vertical Scale Bigger is Better
Vertical scaling makes horizontal scaling more expensive Scaling Quantum
Parallelism is Concurrency Concurrency is not exactly the same na parallelism
Benefits of Threaded Concurrency Sharing Available Resources
Queueing for the GVL Threads Lock On to the Ruby VM
How much concurrency? But you don't have to take my word for it
Amdahl's Law
Job Queueing: Sidekiq Distributed Processing of a Centralized Queue
Latency Based Queueing Latency Bed Scaling
Request Queueing: Puma
Heroku Intelligent Routing Random Routing
Customer Satisfaction
Waiting too long causes frustration
Finding Balance
All Good Things...
Taught by
Ruby Central
