Efficient Mechanism for Enhancing Recovery Performance of SBML in Hierarchical and Distributed Server Architectures

Abstract

The application of the original sender-based message logging (SBML) to the hierarchical architecture may degrade scalability in case of node failure occurrences due to its behavioral limitation. To address this issue, area leader-based mechanism(ALBM) was proposed, but forces all loads of logging and maintaining inter-area messages destined to an area to become concentrated on its leader. In this paper, we propose an efficient mechanism to keep the contents and the receive sequence number of each interarea transmitted message into the area-wide stable storage of its receiver's area. The mechanism allows each recovering node to locally restore to be in a consistent state without any help of the other areas. Also, it allows the logging procedure of each interarea message to be achieved at its own receiver, not the area leader. The experimental results illustrate our mechanism is superior to the previous one, ALBM, in terms of message logging and log information maintenance load.