Analysis of Control Traffic in a Geo-Distributed Collaborative Cloud

Geo-distributed clouds are composed of several virtual machine servers scattered over sites interconnected by a wide area network. In this scenario, the traffic generated by control messages can be prohibitive, because of the high bandwidth costs of these networks. This paper evaluates the impact of this type of traffic in the scalability of the cloud. Based on experiments with the OpenStack orchestrator, we estimate that 100 servers with 15 instantiated virtual machines each send in average 2.7 Mb/s of control traffic to the infrastructure controller. Traffic bursts produced upon the creation and destruction of virtual machines may add to this average control traffic, further aggravating the problem. These results point out that, if the IaaS cloud is poorly designed or the network under provisioned, control traffic can represent a bottleneck especially for small clouds, typically the case of collaborative ones.