Concurrency Overview

Single threaded Concurrency

Each activity requires a separate process
- Has one execution path or "thread" through the program's code
May need "interprocess communication"
- Message queue, pipe, semaphore, shared memory, network socket, etc
Each process has its own private memory space
- Cannot accidentally alter another process's data
Process can be run on different computers over a network
Creating a new process can be slow on some systems
IPC
- Adds complication
- Can be slow to set up
- No direct support in C++

Each thread is an independent execution environment
- Has its own execution stack
- Has its own processor state
Threads are "light weight processes"
- Less overhead to start up
- Smaller task switching overhead
- Easier to share data between threads
All the threads in a process share its memory space
- Can access global data
- Pointers can be passed between threads
Lack of data protection between threads
- Can cause inconsistency and data corruption
- Requires careful programming

Improves responsiveness of the program
- The user is never left staring at a "stuck" program
Improves throughput
- Processing large amounts of data in parallel takes less time
Allows separation of logically distinct operations
- e.g. mail program starts new threads
- Compose an email
- New fetch messages etc
Takes full advantage of modern hardware
- Threads can be distributed among processor cores

Adds complexity to programs
- Code is harder to write and harder to understand
- Bugs are more likely
May not result in faster programs
- Data protection overhead
- Thread coordination overhead
Use only when the benefits outweigh the costs