Revenue management for Cloud computing providers: Decision models for service admission control under non-probabilistic uncertainty

Abstract

Cloud computing promises the flexible delivery of computing services in a pay-as-you-go manner. It allows customers to easily scale their infrastructure and save on the overall cost of operation. However Cloud service offerings can only thrive if customers are satisfied with service performance. Allowing instantaneous access and flexible scaling while maintaining the service levels and offering competitive prices poses a significant challenge to Cloud Computing providers. Furthermore services will remain available in the long run only if this business generates a stable revenue stream. To address these challenges we introduce novel policy-based service admission control models that aim at maximizing the revenue of Cloud providers while taking informational uncertainty regarding resource requirements into account. Our evaluation shows that policy-based approaches statistically significantly outperform first come first serve approaches, which are still state of the art. Furthermore the results give insights in how and to what extent uncertainty has a negative impact on revenue.

Optimal Overbooking Policy for Cloud Service Providers: Profit and Service Quality

“…In the profit distribution process of the cooperative game, we conclude that each member under same situation can gain the same profit in (14); from (15), it shows that is the linear function of, that is, The new game is the direct addition of the original two games after the combination of two independent games. In other words, the cooperative game model in the cloud federation under overbooking policy super additive and is have nonempty core.…”

“…With the growing interest in cloud computing, researchers have begun to focus on enhancing cloud providers' performance and profits. Scholars have studied the problems from different perspectives, e.g., capacity control [5], [15] dynamic pricing [3], [16] [17], service management based on SLA, and client classification [18] to improve profits and resource utilization. In addition, the overbooking policy as a way to improve resource utilization has also been studied.…”

“…As SME users often lack exact information regarding their need for cloud services, they tend to overestimate their resource requirements and reserve more capacity as a buffer [15]. For many services, the peak workload exceeds the average workload, and since few users reserve resources for less than the expected peak, resources are idle at nonpeak times [31].…”

Colocating tasks in data centers using a side-effects performance model

“…Existing models can be roughly categorized into variants of multi-dimensional bin-packing (to model heterogeneous resource requirements), stochastic optimization (to model uncertainty), and statistical approaches. Other issues include pricing and revenue management [Pschel et al, 2015] (in our model, the provider does not select jobs to execute); or distributed scheduling [Sebastio et al, 2017] (our model targets a single data center, rather than a distributed cloud).…”

Optimal Training Parameters and Hidden Layer Neuron Number of Two-Layer Perceptron for Generalised Scaled Object Classification Problem

“…For instance, running 500 eight-core processors takes almost two months to get the function   , , , , p N R B r Q evaluated. This, nonetheless, is realisable owing to cloud services [30], [31] or clusters [32]. …”