Year: 2025
Type: dissertation
Abstract: This master thesis evaluates the performance of Binary, Fibonacci, and Hollow Heaps used as priority queues in Dijkstra's algorithm, with the goal of identifying the most suitable data structure for implementation in Panasonic Avionics' Arc application.The heaps were tested across varying graph types to determine their impact on execution time, memory access patterns, and cache e&ciency.Despite its higher theoretical time complexity, the Binary Heap achieved the shortest execution time for insert in operation benchmarking, owing to its contiguous memory layout and predictable access patterns.The Fibonacci Heap showed the fastest performance for the extract_min and decrease_key operations.Both Fibonacci and Hollow Heaps incurred overhead from pointer usage and irregular memory access.When used as a priority queue in Dijkstra's algorithm on sparse graphs, the Binary Heap completed execution in ".'!-".#(→ the runtime of the Fibonacci Heap, and ".!"-".)(→ that of the Hollow Heap, depending on the graph size.On dense graphs, this performance advantage diminished, and the Fibonacci Heap achieved up to $.!)→ faster execution time than the Binary Heap.This improvement is attributed to the more e&cient decrease_key operation in the Fibonacci Heap, which, in the dense graph setting, outweighs its relatively slower insert and extract_min operations.Implementations in C, Java, and Rust were compared for the Fibonacci Heap, with C showing the overall fastest execution times.Java and Rust o*ered safer abstractions but introduced runtime and complexity trade-o*s.Cache behavior was analyzed using cachegrind, and exhibited that Binary Heap had the fewest cache misses.Prefetching was explored in all implementations; while it showed some bene+t for minimizing L$-misses, overhead added complexity limited its overall execution time.The results demonstrate that the Binary Heap achieves the best execution time and cache performance across the evaluated graphs.
Authors: Tjernström, Klara Juliette, Paulsson, Vera
Language: English
Subfield: Theoretical Computer Science
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7033054206
