Ranking service units for providing and protecting highly available services with load balancing

• Published in: 2010 10th Annual International Conference on New Technologies of Distributed Systems (NOTERE) pp. 33 - 40
• Main Authors: Kanso, A, Khendek, F, Hamou-Lhadj, A, Toeroe, M
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2010
• Subjects: Availability Management Framework; constraints solving; High availability; load balancinge; Load management; Protection; redundancy models
• ISBN: 1424470676; 9781424470679
• ISSN: 2162-1896; 2162-190X
• DOI: 10.1109/NOTERE.2010.5536839

In highly available systems, continuous load balancing in the presence of failure is essential to avoid performance degradation. The Availability Management Framework (AMF) is a middleware service that manages the availability of the applications' services by coordinating their provision among the application's redundant components. The application's components that collaborate to provide a service are logically grouped into a service unit (SU). For management purpose, the SU workload is abstracted as a service instance (SI). Redundant SUs that collaborate to protect a set of Sis are grouped into a service group (SG). The assignment of the Sis to the SUs is a runtime operation performed by AMF. For some redundancy models, NWay and NWayActive, this assignment is performed for each SI according to a predefined ranked list of the SUs in the SG. In this paper we discuss the challenges of defining this ranking at configuration time and propose an approach that automatically generates the ranked list of SUs at that time and guaranties continuous load balancing.