A distinctive component of cloud-based applications is the elasticity control. This component facilitates the adaptation necessary for an application to maintain service quality in the presence of fluctuating demand. Elasticity control achieves this adaptation at runtime by managing the expansion and contraction of resource capacity in response to demand. How to design the rules of elasticity control is a central challenge when deploying cloud-based software. Many application providers express the need to manage the large fluctuations in demand associated with planned events, like marketing events. Existing reactive and predictive elasticity control strategies can be ineffective in managing the surges in workload associated with such planned events. This report will introduce a novel control strategy that integrates expert knowledge about planned events, along with runtime measurements and trend prediction from recent history. We will evaluate how well this strategy can maintain quality of service as planned events alter the load. The initial results presented in this paper are promising and suggest that making an elasticity controller aware of upcoming events is an effective strategy for dealing with event-associated surges in workload.
How many cloud resources does an application provider require to manage workload bursts that often accompany events of public interest, (like product announcements or sporting events), and when will these resources be required? The availability and performance qualities of systems from numerous domains have often been compromised by such bursts,highlighting the importance of these questions. Earlier work begins to address these concerns by presenting burst models, which are parameterized by a single set of burst feature types, to describe bursts that can be associated with different event types from different domains. In this paper we argue that the profile of a burst can differ for different event types, and will depend on a variable number of feature types that describe the burst’s associated event. We contribute a method for creating a workload model that is polymorphic based on event characteristics. Our evaluation uses real world data sets for two different event types, and compares our event-based model to one of the most recent, state of the art models, for workload bursts. Results highlight the dependence of burst profile on the associated event’s definition, and the polymorphic model’s superior accuracy to the other model assessed.
In this paper we extend on earlier work in event aware prediction to provide a generic method for predicting event-associated workload that can be re-used for multiple event types. We propose evaluation methods for this generic event aware prediction approach, which will be carried out in future work, for which a positive assessment can indicate the applicability of event-aware-prediction to a multitude of other domains beyond workload prediction.
Thank you for visiting our website. Would you like to provide feedback on how we could improve your experience?
This site does not use any third party cookies with one exception — it uses cookies from Google to deliver its services and to analyze traffic.Learn More.