Economics of Grids, Clouds, Systems, and Services

The on-demand capability of cloud computing allows consumers to purchase only the computing resources they require, as and when they need it. However, without a view of future demand, cloud providers’ faces challenges in optimising the use of their infrastructure. In this paper, we propose a pricing method for cloud computing which allows providers to schedule virtual machines more efficiently through the use of provision point contracts (PPCs), commonly used for deal-of-the day websites such as Groupon. We show that the model can achieve a reduction of around 2% on the mean number of servers utilised. This may seem a modest percentage, but it can equate to freeing up thousands of physical servers in a single industrial-scale cloud computing data-centre. Additionally, our pricing model prevents discounts being offered where no increase in server efficiency is likely to be achieved. This suggests that the model can be implemented with little risk of it negatively affecting the efficiency of server provisioning. Finally, our results indicate that the cloud-service users who engage with the PPC method can achieve savings of over 20%.

The Cloud computing paradigm offers the illusion of infinite resources accessible to end-users anywhere at anytime. In such dynamic environment, managing distributed heterogeneous resources is challenging. A Cloud workload is typically decomposed into advance reservation and on-demand requests. Under advance reservation, end-users have the opportunity to reserve in advance the estimated required resources for the completion of their jobs without any further commitment. Thus, Cloud service providers can make a better use of their infrastructure while provisioning the proposed services under determined policies and/or time constraints. However, estimating end-users resource requirements is often error prone. Such uncertainties associated with job execution time and/or SLA satisfaction significantly increase the complexity of the resource management. Therefore, an appropriate resource management by Cloud service providers is crucial for harnessing the power of the underlying distributed infrastructure and achieving high system performance. In this paper, we investigate the resource provisioning problem for advance reservation under a Pay-as-you-Book pricing model. Our model offers to handle the extra-time required by some jobs at a higher price on a best-effort basis. However, satisfying these extra-times may lead to several advance reservations competing for the same resources. We propose a novel economic agent responsible for managing such conflicts. This agent aims at maximizing Cloud service provider revenues while complying with SLA terms. We show that our agent achieves higher return on investment compared to intuitive approaches that systematically prioritize reserved jobs or currently running jobs.

With cloud computing, a virtually inexhaustible pool of computing capacity has become available to IT users. However, given the large number of Infrastructure as a Service offers with differing pricing options, the cost-efficient distribution of workloads poses a complex challenge. In this work-in-progress paper, we formally describe the Cloud-oriented Workload Distribution Problem and propose an exact as well as a heuristic optimization approach. Through an evaluation that is based on realistic data from the cloud market, we examine the performance and practical applicability of these approaches.