Copula Analysis of Latent Dependency Structure for Collaborative Auto-Scaling of Cloud Services
Fang Dong, Kui Wu, Venkatesh Bharadwaj Srinivasan, Jianping Wang · 2016
In the field of cloud computing, cloud service composition integrates a group of collaborative sub-services to fulfill a specific business model. Composite cloud service is widely used and is normally implemented by distributed sub-services over the cloud, each modelled as a virtualized function (VF). It is critical that the composite cloud service maintains high Quality of Service (QoS) in the presence of highly-dynamic service requests. The challenge is that each VF has only a myopic view of the whole service process, and scaling up/down individual VF with existing cloud resource auto-scaling strategies does not necessarily lead to better QoS for end users. To solve the problem, this paper proposes an auto-scaling strategy that requires VFs to adjust their cloud resources collaboratively. For collaborative auto-scaling, it is critical to capture the dependence among multiple VFs, and to achieve this goal, we present a novel framework based on copula models to analyze the amount of service calls that may change at different rates for different VFs. We demonstrate how to orchestrate auto-scaling of VFs by predicting future service calls using the temporal dependence captured in the copula model. With real-world trace as well as synthetic data, we demonstrate the benefit of collaborative auto-scaling guided by the copula model.