DVMS: A Proposal to Schedule Virtual Machines in a Cooperative and Reactive Way
Flavien Quesnel · 2014
This chapter presents decentralized approach, distributed virtual machine scheduler (DVMS), to dynamically schedule VMs in data centers. It presents the fundamental concepts behind DVMS such as the distribution of the scheduling work over software agents installed on each node and organized following a ring topology, the triggering of a scheduling procedure when an event occurs on the infrastructure, and the dynamic partitioning of the infrastructure to avoid conflicts between schedulers, without requiring a single service node. The chapter proposes two improvements on the algorithm in charge of processing events. The first improvement enables to speed up the traversal of the ring, by defining on each node of a partition an ever up-to-date shortcut that leads to the first node on the outside of this partition. The second improvement consists of guaranteeing that an event will always be solved if a solution exists.