Hash table and sorted array: a case study of multi-entry data structures in massively parallel systems

The tree, hash table, and sorted array data structures for implementing the primitive operations of a search table are considered. It is argued that the tree structure suffers from the bottleneck problem created by the single entry point, namely, the root, resulting in a linear time complexity. For the hash table and sorted array, the average time complexity for implementing three major operations, namely, insert, delete, and search, is derived. Both analytical and simulation results show that using a sorted array gives a much better performance than using a hash table with linear probing in implementing search table abstraction when the load of the hash table is more than 80%. However, given a hash table having less than 80% load, the average time complexity becomes better than O(log²M), i.e. the hash table gives a better performance in search table implementation than the sorted array. Nevertheless, a larger number of processors is required