Load balancing in cloud-based content delivery networks using adaptive server activation/deactivation

Abstract

Content delivery networks have been widely used for many years providing service for millions of users. Lately, many of these networks are migrating to the cloud for its numerous advantages such as lower costs, increased performance, availability and flexibility in installing new resources. This paper introduces a new approach towards load balancing as well as power reduction in cloud-based content delivery networks allowing for a controlled scaling of operational parameters within the tradeoff between power consumption and quality of experience (QoE). By applying a new proposed adaptive server activation/ deactivation model at each data center in the cloud, unutilized servers at the data center can be switched off to reduce the power consumption. This adaptive model also allows the data center to maintain its performance while being utilized up to 95%. Performance measures of data centers are crucial as there are certain user service level agreements (SLA) for the cloud subscribers that should not be violated; most important is the latency for users' requests. The proposed load balancing algorithm benefits from limited latency for requests by only shifting the load off a data center when it is almost fully loaded. When the load in a data center exceeds a critical level, processes are migrated to underloaded data centers within the cloud. The self-controlled adaptive resource management supports easing of the cloud management.