Journal
M&SOM-MANUFACTURING & SERVICE OPERATIONS MANAGEMENT
Volume 23, Issue 1, Pages 191-209Publisher
INFORMS
DOI: 10.1287/msom.2019.0827
Keywords
capacity expansion; cloud computing; inventory model; dynamic programming
Ask authors/readers for more resources
This study examines the capacity-expansion problem in cloud computing, particularly addressing the disproportionate growth of demand for CPU and RAM. It proposes an expansion strategy based on preconfigured cluster-types and utilizes two algorithms to determine optimal expansion cycle lengths and configurations. The findings provide useful guidance for managing capacity expansion in cloud infrastructure.
Problem definition: The recent surge in demand for cloud services has posed a significant capacity-expansion problem for cloud infrastructure providers. Although the growth of demand for capacity attributes-for example, CPU and RAM-is disproportionate, these attributes are often provided in preconfigured packages (cluster-types), and the fixed ratio of attributes in a package does not match with the time-varying ratio of demand. We analyze a class of expansion policies to determine the timing andmagnitude of expansions, using preconfigured cluster-types, and we examine the optimal configurations of the cluster-types. Academic/practical relevance: Cloud computing is a major technological advance that is influencing businesses significantly, giving rise to an emerging industry but also posing the above-noted capacity-expansion problem. To our knowledge, this is a new issue that has not been studied in the literature. Methodology: We consider growing demand for two attributes and analyze a class of policies that consist of capacity expansion cycles (CECs), whereby capacities are added through sequential or simultaneous replenishments of two configured cluster-types. Results: We first derive the optimal timing and magnitude of expansions for every CEC, and then we devise two algorithms, the dynamic-programming-based (DP) algorithm and the forward-looking (FL) heuristic, to determine the optimal cycle lengths. We also propose a cluster-selection heuristic for choosing the optimal configurations of the cluster-types. Managerial implications: The FL-heuristic is effective, easy to communicate, and can be used as an excellent starting point for the search of the DP-algorithm. Moreover, because there is a desire in practice to reduce the variety of cluster-types, we find conditions under which the employment of only two cluster-types is as efficient as the employment of many cluster-types. Finally, we provide useful guidelines for the optimal configurations of these two cluster-types.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available