Brief announcement: tree decomposition for faster concurrent data structures

We show how to partition data structures representable by directed acyclic graphs, i.e. rooted trees, to allow for efficient complex operations, which lie beyond inserts, deletes and finds. The approach potentially improves the performance of any operation modifying more than one element of the data structure. It covers common data structures implementable via linked lists or trees such as sets and maps. We demonstrate its simplicity and its effectiveness using a concurrent sorted linked list. We achieve a speedup of up to 250% even for small divisions. Categories and Subject Descriptors: E.1 Data structures General terms: algorithms, performance Key Words: concurrent data structures, graphs

• Type: Conference
• Year: 2010
• Where: PODC
• Authors: Johannes Schneider, Roger Wattenhofer
• Keywords: Data Structures, Directed Acyclic Graphs, Distributed And Parallel Computing, Linked List, PODC 2010