Peer-to-peer bargaining in container-based datacenters

In container-based datacenters, failure-prone components are sealed in pre-packaged shipping containers, and component failures over time reduce the availability of resources. From the perspective of services, application instances can usually be migrated across the boundary of containers as virtual machines (VMs). In such an environment, it would be sometimes beneficial to migrate application instances to take better advantage of available resources. In this paper, we first point out that application placement strategies that are singularly focused on the efficiency of resource usage may not be beneficial, especially when resources are over-provisioned in container-based data centers. We believe that application instances should be placed based on the Buffet principle, using resources in an aggressive fashion. Once failures occur, application instances can be migrated across containers, by allowing containers to bargain with each other in a peer-to-peer fashion, treating application instances with different resource requirements as "commodities" in a Nash bargaining game. We show that the ratio of utilizing resources can improved by establishing such local Nash bargaining games, where two containers bargain and trade VMs with each other to increase their own utilities. With simulation, we verify the effectiveness of the proposed bargaining games.