Abstract

The global provisioning of some types of evolved services over the Internet requires the deployment of servers; the approach is usually called service overlays. An optimal design of the number of servers and of their location is critical in order to minimize the deployment cost while providing the quality of service expected by users.

In this paper, we propose an original modelling, analysis and optimization approach based on stochastic geometry to deal with this design issue. Two particular cases of service overlays are emphasized: servers deployed for content delivery (i.e. to deploy a content delivery network or CDN) and servers deployed for reliable multicast communications.

In our approach, the network topology, the subscribers’ location and the servers’ location are represented by point processes in the plane. Point process theory and Palm Calculus are used in order to obtain closed form expressions for the expected value of realistic cost functions. These closed forms allow us to evaluate the optimal cost of a future deployment.

Keywords: Network planning; Stochastic geometry; Content delivery network; Multicast transport

Article Outline

1. Introduction

2. Optimal content delivery networks deployment

2.1. Context and motivations

2.2. Presentation of the model

2.3. Cost function

2.4. Extended model

2.5. Computation

2.6. Example

3. Optimization of overlays for reliable multicast transport

3.1. Technical presentation: Multicast

3.2. Technical presentation: Reliable multicast transport protocols

3.3. Model

3.3.1. Modelling users’ and servers’ location

3.3.2. Cost functions

3.3.3. Loss model

3.3.3.1. Loss model on the data channel

3.3.3.2. Loss model on the control channel

3.4. Computation

3.4.1. Preliminary results

3.4.2. Mean number of retransmissions from the higher level

3.4.3. Mean number of retransmissions between the levels H−1 and H−2

3.4.4. Mean number of retransmissions in the inferior levels (within the clusters)

3.5. Example

4. Conclusion

References

Vitae