Network Traffic Analysis-based Service Scaling Algorithms for Cloud Data Center

Abstract

Today, various types of cloud-based services are available on the Internet such as e-banking, social networking, e-commerce, online game, and registration systems. Users access those services from any location using a computer or mobile device with the help of deployed applications on the cloud data centers. However, providing scalable service in cloud data center is still challenging issue, since number of cloud users is increasing exponentially and it also stores sensitive and proprietary data. Thus, this dissertation deals with the problems of service scaling in cloud data center.

In the first part of this dissertation, we consider traffic-aware virtual machine (VM) placement scheme in software-defined data center (SDDC). In conventional strategies, the communication dependency between VMs is not considered when a service is expanded by creating additional VMs. Thus, highly dependent VMs may be placed in different physical machines (PM), which will significantly increase the data traffic between PMs. This will negatively impact the service performance because the data traffic between the service and the customer flows through the same physical network. In this regard, it is desirable to minimize the traffic volume in physical network to improve the service performance. To reduce the amount of data traffic in the underlying network and improve the service performance, we propose a traffic-dependency-based strategy for VM placement in SDDC.

In the second part of this dissertation, we consider how to efficiently schedule the migration order of multiple VMs in order to reduce service downtime during cloud-to-cloud (C2C) migration. If some VMs belong to the same service, it is important to schedule their migration in short time intervals in order to decrease the service downtime. However, it is a critical issue to decide migration order of the multiple VMs, provided that a cloud operator is unavailable to distinguish among the various dependent VMs due to the complexity of service deployment. If the VMs are used for multiple services, the cloud operator would not be able to identify the dependency of the VMs. For solving the issue, we propose a service-aware strategy for C2C migration of multiple VMs, which analyzes the dependency between the VMs, using network traffic intensity to determine the migration sequence of dependent VMs in order to decrease the service downtime. The proposed method that exploits the dependency among the VMs significantly reduces the service downtime, while the service downtime increases exponentially, when the VMs are migrated randomly