Connecting Wide-Area Cloud Resources with Virtual Networking

CloudNets

Internet Network Architectures (INET)
TU Berlin and Deutsche Telekom Lab
Contact: Stefan Schmid
VNet Embeddings

News

Project Overview

CloudNets are (wide-area) virtual networks (VNets) connecting geographically distributed cloud resources (e.g., large or small datacenters). The network virtualization paradigm allows to run multiple CloudNets on top of a shared physical infrastructure. These CloudNets can have different properties (provide different security or QoS guarantees, run different protocols, etc.) and can be managed independently of each other. Moreover, (parts of) a CloudNet can be migrated dynamically to locations where the service is most useful or most cost efficient (e.g., in terms of energy conservation). Depending on the circumstances and the technology, these migrations can be done live and without interrupting ongoing sessions. The flexibility of the paradigm and the decoupling of the services from the underlying resource networks has many advantages; for example, it facilitates a more efficient use of the given resources, it promises faster innovations by overcoming the ossification of today's Internet architecture, it simplifies the network management, and it can improve service performance.

We are currently developing a prototype system for this paradigm (currently based on VLANs), which raises many scientific challenges. For example, we address the problem of where to embed CloudNet requests (e.g., see [1] for online CloudNet embeddings and [2] for a general mathematical embedding program), or devise algorithms to migrate CloudNets to new locations (e.g., due to user mobility) taking into account the corresponding migration cost (see [3] for single server migration, [4] for multi-server allocation, and [5] for migration in a multi-provider network). Since the arrival of CloudNet requests or the demand for a certain service may be hard to predict, we often employ the tools of competitive analysis to devise online algorithms that do not rely on any predictions about the future but still guarantee a provable worst-case performance.

The project also has interesting economic dimensions (e.g., regarding pricing mechanisms).

People

This project is a joint work with NTT DoCoMo Eurolabs Munich (group headed by Wolfgang Kellerer).

Current project lead at INET:

Anja Feldmann, Carlo Fürst, Johannes Grassler, Arne Ludwig, Matthias Rost, Stefan Schmid

Magazine Articles

Publications

Embeddings:

Migration:

Prototype architecture and economics:

Software-Defined Networking (SDN):