PDS1 - Paralelné a distribuované systémy

1. exercise

• Concurrent Execution in Python, Speed Up Your Python Program With Concurrency
• kód z cvičenia (html)

2. exercise

• Bounded Semaphore, Multiple consumers and producers
• code (html)

3. exercise

• Multiprocessing, SharedMemory
• code (html)
• HW: Create program computing nansum with SharedMemory and without. Compare running time. (BONUS: memory)

4. exercise

• Fibonacci numbers, Primes with Cython
• Mandelbrot with Numba
  Tasks
  1. complex numbers in Python
  2. complex numbers in Cython
  3. complex numbers with JIT in Numba
  4. complex numbers no sqrt, JIT in Numba
  5. real numbers, noq sqrt, JIT in Numba
  6. guvectorize complex64+complex128
  7. cuda@numba (CUDA-capable grafic card, Colab, or on server .56 from cuda tasks), try different dimensions blockdim/griddim
• code (html)

7. exercise

• Tasks
  1. Compute the sum of array - several solutions and find biggest speed-up.
     from random import randrange
     arr = [randrange(10, 1000) for _ in range(100_000_000)]
  2. Compute the sum of two arrays (e.g. [1, 2] + [3,4] = [4, 6]) - several solutions and find biggest speed-up.
  3. Compute the multiplication of two matices - several solutions and find biggest speed-up.
• code (html)

10. exercise MPI

• MPI tutorial
• TASKS (run on the server pds_run taskX.py >taskX.out):
  1. task1.py: Every process generates two numbers, reduce minimums to the rank 1.
  2. task2.py: Every process generates one number, compute the MPI_Scan and explain expected results in source file.
  3. task3.py: Every process generates one number, compute the MAXLOC to all processes.
• code (html), codes (server)

11. exercise MPI

• Other materials: MPI course material, Parallel Programming in MPI and OpenMP
• generating in Python
• HW: Implement solution of 0-1 Knapsack problem
  1. Master generates prefixes and distribute them to workers. Worker calculates all solutions for given prefix. At the end worker sends his best solution. Master prints the maximum value.
  2. Show also the solution (which items you choose), not only the maximum value - you can use MAXLOC with INT2 or communicate with all workers.
  3. Run on server pds_run hwMPI2.py >hwMPI2.out.
• code (html)

© 2003-2025