Glossary · CS330

Core parallel computing#

Baseline: The reference used for comparison. In Phase II, the baseline is the 1-thread run.
Benchmark: A controlled performance test that measures execution time under fixed conditions.
Concurrency: Multiple tasks making progress in overlapping time periods. Concurrency does not always mean the tasks run at the same instant.
Correctness check: A step that confirms the multithreaded result matches the single-thread result.
Data parallelism: A strategy that applies the same operation to different parts of the data at the same time.
Execution time: The total time a program or configuration takes to finish. This project reports it in milliseconds.
Overhead: The extra cost of running in parallel, such as thread creation, synchronization, scheduling or process management.
Parallelism: Using multiple processing units, threads or processes to do work simultaneously.
Scalability: How performance changes as more threads or processes are added.
Speedup: How much faster a parallel version is than the baseline. In this project, Speedup = T(1) / T(n).
Synchronization: The coordination that ensures parallel tasks finish safely and their results are combined correctly.
Throughput: The amount of work completed per unit of time. This project focuses on execution time, which is closely related.

Child process: A process created by another process with fork().
fork(): A Unix system call that creates a new child process. Phase I uses it to divide matrix multiplication across processes.
wait(): A system call the parent uses to pause until its child processes finish.
Matrix multiplication: The Phase I workload: multiplying matrices A and B to produce matrix C.
Process: An independent running program with its own execution state and resources.
Process-based parallelism: Parallel execution achieved by running multiple processes rather than multiple threads.
Virtual machine (VM): A software-based computer. Phase I ran Ubuntu Linux in one to keep the environment controlled.

Array partitioning: Dividing the input array into non-overlapping ranges so each thread processes a different slice.
Chunk: A contiguous part of the array assigned to one worker thread.
join(): In Java, makes one thread wait until another finishes. Phase II calls it before combining the partial sums.
Multithreading: Using multiple threads inside one program to do work in parallel.
Partial sum: The result one worker computes locally, before all the workers' results are added together.
Runnable: A Java interface for code that can run on a thread. Summer.java implements it.
Sum of cubes: The Phase II workload: each array value contributes x^3 to the final result.
Thread: A lightweight unit of execution inside a process. Threads share the same memory but can work on different parts of a task concurrently.
Thread configuration: The number of threads used in one experiment run: 1, 2, 4, 6 or 8.
Worker thread: A thread responsible for computing one assigned part of the input.

Average execution time: The mean runtime across repeated runs of the same configuration.
Configuration: A specific test setup, such as a given matrix size, process count or thread count.
Controlled experiment: An experiment where only the chosen variables change and everything else stays fixed.
Graph interpretation: An explanation of what a graph shows and why the trend appears.
Machine specifications: The hardware and software details of the benchmark machine, such as CPU, RAM, OS and Java version.
Percentage improvement: The percentage reduction in execution time compared with the baseline: ((T1 - Tn) / T1) * 100.
Raw timing data: The run-by-run execution times, before averaging.
Reproducibility: Whether the same setup and method produce comparable results when the experiment is repeated.

No terms match that filter.