A QoS-aware self-correcting observation based load balancer

• Published in: The Journal of systems and software 2016-05, Vol.115, p.111-129
• Main Authors: Chandakanna, Veerabhadra Rao, Vatsavayi, Valli Kumari
• Format: Article
• Language: English
• Subjects: Benchmarks; Cluster analysis; Clusters; Computer programs; Distributed processing; Dynamic load balancing; Estimates; Fault tolerance; Monitoring; Monitors; Quality of Service; Redundant; Software; Studies
• ISSN: 0164-1212; 1873-1228
• DOI: 10.1016/j.jss.2016.01.042

•Redundant QSLBs collaborate to estimate servers’ capabilities, and share the capacity.•The QSLB allows to set the QoS benchmarks, and monitor the QoS parameters.•ARSM and NRPM models estimate the cluster capacity needed to improve the QoS.•Different architectures to implement the QSLB are explored and evaluated.•Experiments were conducted to test and analyze the QSLB features. Any service offered by a load balanced cluster is deployed on every member of the cluster. The Sliding window based Self-Learning and Adaptive Load Balancer (SSAL) is an observation based load balancer that optimizes throughput. It gives single point entry to access any service hosted on the cluster. This paper proposes a QoS-aware and Self-correcting observation based Load Balancer (QSLB) that extends the SSAL to (i) prevent the single point of failure of the load balancer, (ii) manage the cluster capacity, (iii) support the QoS monitoring, and (iv) estimate the cluster capacity needed to meet the QoS benchmarks. Redundant QSLBs collaborate to estimate the capabilities of the individual cluster members, share the available cluster capacity, and monitor the QoS parameters. Two models to estimate the cluster capacity needed to meet the QoS benchmarks are proposed. Experiments were conducted to test the QSLB’s features. The experimental results confirmed that (i) the overhead to support these QSLB features is minimal, (ii) the QSLBs retained their share of the cluster capacity even in dynamic environments, and (iii) using the recommended cluster capacity improved the QoS met percentage. The proposed model improves fault tolerance, assists in cluster capacity management, and monitors the QoS.