Parallel Algorithm Development

Question 1

What is data decomposition?

Answer 1

The data on which the computations are performed is partitioned. This turns a single task across a lot of data into a single task across many smaller chinch’s of data

Question 2

What is task decomposition?

Answer 2

By splitting the tasks into independent task groups. These groups of task can be done without interference of another processor/thread.

Question 3

What is loop decomposition? And what is a challenge faced when implementing loop decomposition?

Answer 3

It involves splitting a loop between threads/processors by allocating loops in chunks or in a round robin fashion to each thread/processor. An issue that can be run into is with data decencies i.e. a loop has a decency on the outcome of a previous loop

Question 4

How does a forall loop parallelism type work?

Answer 4

Each assignment is treated like as a separate array assignment and thus uses up to date assignment of previous and subsequent array values.

Question 5

How does a dopar for loop parallelism work?

Answer 5

Updates in one iteration are not visible to other iterations and therefore values of array are equal to the initial assignment before the loop has started