Melhorando o balanceamento de carga em ambientes virtualizados usando a correlação de Pearson
Moraes Filho, Nilson Rubens de
MetadataMostrar registro completo
Virtualization is one of the foundations of cloud computing as it allows better utilization of computing resources in a data center. There are different virtualization approaches that offer similar functionality, but with different levels of abstraction and methods. In this sense we can mention the use of Virtual Machines and containers. We define Virtual Element (VE) as a virtual machine or a container, and we will use this concept to make our load balancing approach generic. Load balancing can be achieved through live migration of VEs, reducing energy consumption, enabling better distribution of computational resources and allowing customers to move VEs from a cloud provider to another one that may offer better SLA or costs. There are some methods that addresses the load balancing improvement in a data center. One of them applies the Pearson correlation coefficient related to CPU usage, to migrate VEs from an overloaded host to another that have better availability of resources. The Pearson correlation coefficient estimates the dependency level between quantities. Yet, live migration is a virtualization feature that allows a VE to be transferred from one equipment to another, keeping the active processes running. However, to migrate a VE that has strong dependency with the internal network traffic from a host, can create an increase in the overall network consumption due to the migration of the VE to another server, topologically distant from the current host. This dissertation defines a heuristic that has as objective improve the migration decision process of VEs. The heuristic applies Pearson's correlation coefficient and takes in consideration not only CPU consumption, but also the internal network traffic between VEs. Results shown that the application of the heuristic improved the decision process in at least 18% compared to a method that considers only CPU correlation coefficient.