Generalized and Resource-Efficient VNet Embeddings with Migrations

Abstract

This paper attends to the problem of embedding flexibly specified CloudNets, virtual networks connecting cloud resources (such as storage or computation). We attend to a scenario where customers can request CloudNets at short notice, and an infrastructure provider (or a potential itermediate broker or reseller) first embeds the CloudNet fast (e.g., using a simple heuristic). Later, however, long-lived CloudNets embeddings are optimized by migrating them to more suitable locations, whose precise definition depends on a given objective function. For instance, such migrations can be useful to reduce the peak resource loads in the network by spreading CloudNets across the infrastructure, to save energy by moving CloudNets together and switching off unused components, or for maintenance purposes. We present a very generic algorithm to compute optimal embeddings of CloudNets: It allows for different objective functions (such as load minimization or energy conservation), supports cost-aware migration, and can deal with all link types that arise in practice (e.g., full-duplex or even wireless or wired broadcast links with multiple endpoints). Our evaluation shows that such a rigorous optimization is even feasible in order to optimize a moderate-size CloudNet of full flexibility (e.g., a router site, a small physical infrastructure or virtual provider network).