Proactive Event Aware Cloud Elasticity Control

Abstract

An event that can attract the interest of a large internet populace, can bring about large bursts in demand for event-associated resources. This phenomenon is notorious for degrading the QoS of numerous web-applications overwhelmed with workload bursts that accompany events like online ticket releases, sales events and result announcements. Such reduced QoS can translate to reduced revenue and degraded reputation for application providers. While cloud computing has the potential to alleviate this problem by offering computing resources on an as-needed basis to match event-associated demand, important challenges remain in finding the right resource control strategies to cost-effectively scale resources, and in overcoming the initialisation lag associated with resource acquisition. This thesis recognises an implicit link between events and bursts, and introduces event-aware burst modelling and burst prediction techniques that can forecast the time, shape, magnitude and length of event-associated workload bursts. We show how these techniques can assist application providers in making more informed decisions about when to acquire and release resources to match event-associated demand, in order to maintain application QoS during event-associated bursts.