Project 3: Parallel Cryptocurrency Miner

See instructions here: https://www.cs.usfca.edu/~mmalensek/cs220/assignments/project-3.html

Written Responses

Edit this readme.md file to answer the following questions:

Busy-waiting vs. Condition Variables

Once you have condition variables working in your program, compare the performance of busy waiting with condition variables. To ensure a fair comparison, use a single thread.

Report for the original program as well as modified version:

Argument: ./mine 1 24 'Hello CS 220!!!'

Condition Variables:

1. Total run time: 1.85s
2. Hashes per second: 547563.49
3. CPU usage: 112.8%

Busy Waiting:

1. Total run time: 1.75s
2. Hashes per second: 576781.53
3. CPU usage: 199.7%

Which version performs better? Is this the result you expected? Why or why not?

In terms of speed, the busy waiting version performed better. I expected that to be the case due to the fact that busy waiting moves at a faster pace than condition variables since condition variables requires more communication. However, in terms of CPU usage conditional variables performed better than busy waiting. Again, I expected this to be the case since busy waiting inefficiently uses CPU to loop while waiting to continue as opposed to condition variables where minimal CPU is used to wait to move on.

If the problem were to scale up and a great number of threads are used, it is obvious why condition variables is the safer way to go.

Nonces Per Task

Experiment with different values for NONCES_PER_TASK. What value yields the best performance in terms of hashes per second on your machine?

120 seems to be the best NONCES_PER_TASK for the best results.

Performance I

When evaluating parallel programs, we use speedup and efficiency. Why are these metrics not as useful when measuring the performance of our parallel crytocurrency miner?

With parallel programs, the speedup and efficiency is effected due to how much the workload is split up. However, with threads that is not the case. In the case of the cytocurrency miner, the more threads leads to more consumers which leads to the producer able to produce work at a faster pace until hitting a point of diminishing returns.

Performance II

Using any of the kudlick machines (in our 220 classroom), what is the highest performance you were able to achieve in terms of hashes per second? What configuration did you use (NONCES_PER_TASK, number of threads, block data, compiler options)?

• Hashes per second: 4531938.64
• NONCES_PER_TASK: 120
• Number of threads: 5
• Block data: 'Hello CS 220!!!'
• Compiler options: -O3