4.2

4.2 Multicore Programming

Concept of Multicore Systems:

• Historical Evolution: Single-CPU systems evolved into multi-CPU systems for better performance. The modern trend is placing multiple cores on a single chip.

• Multicore Systems: A multicore system has multiple computing cores, each appearing as a separate CPU to the operating system.

• Concurrency vs. Parallelism:

  • Concurrency: Multiple tasks making progress over time (e.g., tasks interleaved on a single-core system).

  • Parallelism: Multiple tasks executed simultaneously (e.g., tasks running on separate cores).

4.2.1 Programming Challenges for Multicore Systems

Key Challenges in Multicore Programming:

1. Identifying Tasks: Breaking applications into separate tasks that can run concurrently or in parallel on different cores.

2. Balance: Ensuring tasks are evenly distributed in terms of work and value to avoid inefficiencies in parallel execution.

3. Data Splitting: Dividing data appropriately for each task to run on separate cores without bottlenecks.

4. Data Dependency: Managing dependencies between tasks, ensuring proper synchronization when one task depends on another's data.

5. Testing and Debugging: The complexity of testing/debugging increases with parallel execution due to multiple possible execution paths.

Impact on Software Development:

• The rise of multicore systems requires rethinking traditional software design.

• Software developers and educators must focus on parallel programming and efficient use of multiple cores.

4.2.2 Types of Parallelism

Two Main Types of Parallelism:

1. Data Parallelism:

   • Involves distributing subsets of the same data across multiple cores.

   • Example: Summing elements of an array on a multi-core system. One thread sums the first half of the array, while another sums the second half.

2. Task Parallelism:

   • Involves distributing distinct tasks (threads) across multiple cores. Each thread performs a unique operation.

   • Example: Two threads performing different statistical operations on the same array of data.

Hybrid Parallelism:

• Applications may use both data parallelism and task parallelism in a combined approach for optimal performance.

Summary of Multicore Programming:

• Multicore programming is essential for leveraging modern hardware.

• It involves challenges related to task identification, workload balance, data splitting, managing dependencies, and debugging.

• Parallelism can be data-based (distributing data across cores) or task-based (distributing tasks across cores).