Learning-Based Data Envelopment Analysis for External Cloud Resource Allocation

A mature cloud system needs a complete resource allocation policy which includes internal and external allocation. They not only enable users to have better experiences, but also allows the cloud provider to cut costs. In the other words, internal and external allocation are indispensable since a combination of them is only a total solution for whole cloud system. In this paper, we clearly explain the difference between internal allocation (IA) and external allocation (EA) as well as defining the explicit IA and EA problem for the follow up research. Although many researchers have proposed resource allocation methods, they are just based on subjective observations which lead to an imbalance of the overall cloud architecture, and cloud computing resources to operate se-quentially. In order to avoid an imbalanced situation, in previous work, we proposed Data Envelopment Analysis (DEA) to solve this problem; it considers all of a user’s demands to evaluate the overall cloud parameters. However, although DEA can provide a higher quality solution, it requires more time. So we use the Q-learning and Data Envelopment Analysis (DEA) to solve the imbalance problem and reduce computing time. As our simulation results show, the proposed DEA+Qlearning will provide almost best quality but too much calculating time.