ECONOMICALLY EFFICIENT CACHE FRIENDLY ADJACENCY LIST FOR GRAPH PROCESSING ALGORITHMS

Abstract

Graphs are a fundamental part of computer science and various scientific fields. They are used in numerous applications, from social networks to electronic design automation. Efficient graph representation is important for modern algorithms, especially in economical contexts, where resource-efficient implementations play a significant role by reducing the resource consumption. This is essential in environments where computing resources are a high priority. There are many representations of graphs as data structure. One of these types is adjacency list: it is collection of linked lists which shows the connection of source and destination vertices. Adjacency lists are more memory-efficient than adjacency matrices because they only allocate memory for vertices with at least one edge in the graph. However, traditional implementations of adjacency lists do not fully use the features provided by modern CPUs, such as the cache line. Our design of Adjacency List is cache friendly and is keeping the main structure of traditional implementation. The main idea is derived from C#’s dictionary implementation for hash tables, which uses one array combining all data that should be in linked lists. Our results show that cache friendly implementation of Adjacency list outperforms it in lookup and insertion operations.