Dependability evaluation of the spread/distributed autonomous replication management framework in a realistic deployment scenario

Abstract

DARM is a middleware framework for managing and distributing services through replication. It is based on the underlying group communication framework Spread, which delivers fast and reliable group communication, with membership management, where DARM use multiple separate groups to separate different areas of functionality. DARM is designed to tolerate node faults and network partitioning, while ensuring availability and reliability to multiple services on each network segment through recovery procedures. In this thesis is presented an approach to determine dependability attributes of a fully deployed DARM system, with multiple realistic services included. Also presented is the work that has been done to prepare and extend the DARM framework with toolskits that are considered indispensable to distributed services. DARM is compared with similar solutions, and presented as a potential framework for acting as a cloud computing middleware, supporting both cloud infrastructure and service development. Results of this thesis indicate that DARM is reliable, performing consistently with high grades of availability and MTBF. Remaining defficiencies are reduced to acceptable levels.