Concurrency
in Go
Agenda
General Concepts
- Threads vs. Green Threads
- Concurrency & Parallelism
Goroutines
- Implementation, Channels
- Concurrency Patterns
Threads
Smallest sequence of programmed instructions that can run independently
Managed by OS
- Userspace
- Kernel Space
Threads
Smallest sequence of programmed instructions that can run independently
Managed by OS
- Userspace
- Kernel Space
Context Switch by swapping out data at CPU level
- Expensive Process
Threads
C10k is the problem of concurrently handling ten thousand connections.
Handling many concurrent connections is a different problem to handling many requests per second.
The latter requires high throughput
(processing them quickly)
While C10K does not have to be fast,
but requires efficient scheduling of connections.
Green Threads or Coroutines
Smallest sequence of programmed instructions that can run independently
Smallest sequence of programmed instructions that can run independently
Managed by
- Program Runtime
- Virtual Machine
Context Switch by swapping out data at runtime level
- Lightweight process
Green Threads or Coroutines
Threads
- Managed by OS
- Hardware level
- Expensive Process
- Example: Apache Server
Coroutines
- Managed by Runtime
- In Program
- Lightweight Process
- Example: Nginx
Threads
- Managed by OS
- Hardware level
- Expensive Process
- Example: Apache Server
Coroutines
- Managed by Runtime
- In Program
- Lightweight Process
- Example: Nginx
- Java
- C/Cpp
- Node/Javascript
- Python
- Ruby
Threads
- Managed by OS
- Hardware level
- Expensive Process
- Example: Apache Server
Coroutines
- Managed by Runtime
- In Program
- Lightweight Process
- Example: Nginx
- Java
- C/Cpp
- Node/Javascript
- Python
- Ruby
I/O Bound
vs.
CPU Bound
CPU Bound
Processor is the only component being used for execution
I/O Bound
Execution is dependent on the input-output system and its resources, such as disk drives and peripheral devices
Boundaries define if a program is sequential or concurrent/parallel
Tasks
- Call
- Dinner
Tasks
- Email
- Write Email
- Review/Send Email
- Call
- Person 1
- Person 2
- Dinner
- Prepare Food
- Eat Food
Concurrency vs. Parallelism
Call 1
Write Email
Prep Food
Call 2
Send Email
Eat Food
You
Concurrency
Concurrency vs. Parallelism
Call 1
Write Email
Eat Food
You
Call 2
Prep Food
Eat Food
Friend
Call 3
Send Email
Parallelism
Single Threaded Concurrency
- Only one task at a time
- Runs on a single core
- Uses an Event Loop
Call 1
Email 1/2
Prep Food
Call 2
Email 2/2
Eat Food
Event Loop
Pop Quiz
Parallelism
Parallelism + Event Loops = Goroutines
- Tasks executed parallelly == number of CPU cores
- Uses modified/extended event loop
Call 1
Email 1/2
Prep Food
Call 2
Email 2/2
You: Core #1
Friend: Core #2
Global
Run
queue
Goroutines
- # of Event Loops = # of processors
- Boundaries using `go` keyword
- Boundaries on I/O calls
Pop Quiz
Goroutines (Rabbit Hole)
- Go runtime/scheduler will switch active goroutine every 10ms
- N:M threading model
- Work Stealing Scheduler
- Hoare's Communicating Sequential Processes (CSP)
Concurrency Patterns
Do not communicate by sharing memory; instead, share memory by communicating.
Communicate via Channels
See Sample 7
Channels for concurrency
- Unbuffered Channels to Synchronize
- Buffered Channels to Parallelize
Fan-Out, Fan-In
See Sample 9