More Related Content
Similar to Concurrency vs parallelism (20)
Concurrency vs parallelism
- 1. Concurrency vs Parallelism Concurrency is NOT parallelism.
- 2. SJ Campus Model Speed up from 4 minutes to 40 seconds (6 times faster) If also include Optimisation time, we speed up around 12 times.
- 3. Concurrency vs Parallelism • Rob Pike, the inventor of Go and member of Unix team • Gave a talk in 2013, when introducing the Go concurrency primitives • “Concurrency is NOT parallelism” • https://blog.golang.org/waza-talk
- 4. Concurrency vs Parallelism • The modern world is parallel • CPU, Network, Users, Workers, etc • A lot of programming language provides concurrency primitives • Concurrent execution • Synchronisation and messaging
- 5. Concurrency vs Parallelism • Concurrency • Programming as the composition of independently executing processes. • Parallelism • Programming as the simultaneous execution of (possibly related) computations.
- 6. Concurrency vs Parallelism • Concurrency is about dealing with lots of things at once. • Parallelism is about doing lots of things at once. • Concurrency is about structure, parallelism is about execution. • Concurrency provides a way to structure a solution to solve a problem that may (but not necessarily) be parallelizable.
- 7. Concurrency plus communication • Concurrency is a way to structure a program by breaking it into pieces that can be executed independently. • Communication is the means to coordinate the independent executions. • CSP (Communicating Sequential Processes), 1978 • https://www.cs.cmu.edu/~crary/819-f09/Hoare78.pdf
- 8. Parallelism example • Multi-Threading • Multi-Processing • Clusters
- 9. Parallelism example Concurrent Dictionary (Meshes, …) Thread 1: IO, Read and parse data from disk. Thread 2: loop forever and if the mesh does not have collider, attach it. Thread 3: loop forever and if there is new mesh instances, merge it. 4 threads used including main thread.
- 10. Problem? Concurrent Dictionary (Meshes, …) Thread 1: IO, Read and parse data from disk. Thread 2: loop forever and if the mesh does not have collider, attach it. Thread 3: loop forever and if there is new mesh instances, merge it. Workload: High Workload: Medium, but need wait for main thread Workload: low Low-level concurrency model is NOT able to utilise the computational resources
- 11. Problem? Concurrent Dictionary (Meshes, …) Thread 1: IO, Read and parse data from disk. Thread 2: loop forever and if the mesh does not have collider, attach it. Thread 3: loop forever and if there is new mesh instances, merge it. Loop forever, busy wait Workload: low Loop forever, busy wait Low-level concurrency model creates OVERHEAD
- 12. Modern Concurrency Model • Task (promise, coroutine, worker) model • Thread pool model • Be able to utilise all computational resources • Be able to write code in sequential way
- 13. Task Model Concurrent Queue (Tasks…) Push task to the queue Thread 1: loop forever and try dequeue, if there is a task, executes it. Thread 2: loop forever and try dequeue, if there is a task, executes it. Thread 3: loop forever and try dequeue, if there is a task, executes it. Worker Threads Thread Pool
- 14. Problem? Concurrent Queue (Tasks…) Push task to the queue Thread 1: loop forever and try dequeue, if there is a task, executes it. Thread 2: loop forever and try dequeue, if there is a task, executes it. Thread 3: loop forever and try dequeue, if there is a task, executes it. Worker Threads When thread 1 is busy, thread 2 can take the task and execute Be able to utilise all resources (if there are tasks in the queue)
- 15. Problem? Concurrent Queue (Tasks…) Push task to the queue Thread 1: loop forever and try dequeue, if there is a task, executes it. Thread 2: loop forever and try dequeue, if there is a task, executes it. Thread 3: loop forever and try dequeue, if there is a task, executes it. Worker Threads If there is no tasks or just a few tasks in the queue, we can reduce the amount of the worker threads. And increase them when more tasks come in. (Less overhead)
- 16. Dependency? Concurrent Queue (Tasks…) Push task to the queue Thread 1: loop forever and try dequeue, if there is a task, executes it. Thread 2: loop forever and try dequeue, if there is a task, executes it. Thread 3: loop forever and try dequeue, if there is a task, executes it. Worker Threads What if tasks have dependency? i.e. task 2 must execute after task 1. We can enqueue task 2 after task 1 finishes.
- 17. Priority? Concurrent Queue (Tasks…) Push task to the queue Thread 1: loop forever and try dequeue, if there is a task, executes it. Thread 2: loop forever and try dequeue, if there is a task, executes it. Thread 3: loop forever and try dequeue, if there is a task, executes it. Worker Threads We can use multiple task queues if task has priority. Or we can create a special task queue for main thread. Concurrent Queue (Tasks…) Concurrent Queue (Tasks…)
- 18. Question • Unity StartCoroutine • Is it a concurrency model? • Does it run parallel? • How does Unity implement the coroutine? • What is the potential issue of the Task model?