CSC $240$ Exercise $5$ Specifications

CSC $240$

Exercise $5$

$5\mathrm{.}1$ Run the parallel merge sort on your own computer, trying various chunk sizes, to see how much

speed$-$up you can obtain from the available parallelism. Write a report of at least $100$ words.

$5\mathrm{.}2$ The C$++$ thread library provides a function that returns the number of threads that the

hardware is capable of running. Here is how to get the number of threads available:

unsigned int maxthreads $=$ thread::hardware$\_$concurrency $()$;

Modify the parallel merge sort so that it will utilize the number of threads available $($maxthreads$).$

Calculate the maximum chunk size allowed to equally distribute the job amongst the threads:

MAX$\_$ITEMS$/$maxthreads$.$ Make sure the chunk size can't be greater than this ratio, but it can be

less. Record the results from this modification.

$5\mathrm{.}3$ Comparing the results of your experiments in Exercise $5\mathrm{.}1$ with the performance obtained from

the approach in Exercise $5\mathrm{.}2,$ does using a thread count that is greater than the number available

from the hardware ever produce faster performance? Is it possible to get consistently better

performance on average, or is the overall performance just within the normal range of variation?

Will chunk sizes slightly less than or slightly greater than the chunk sizes used in Exercise $5\mathrm{.}2$ yield a

better overall performance?

$5\mathrm{.}4$ Now, redo $5\mathrm{.}1$ to $5\mathrm{.}3$ on another system that allows more or less threads to run concurrently.

Write a report of at least $250$ words that identifies your observations and conclusions regarding this

experiment.