Distributed resource allocation for virtual networks

The recent evolution of network infrastructures allows for more elasticity and dynamicity to network management. The 5th generation of networks (5G) allows the creation of Chains of Network Functions on top of virtual networks (”slices”) that can be multi-domain, or even multi-operators. Centralised solution usually used for network management might not be adequate for these newer use cases. This thesis makes the case for the opportunity of a distributed solution to the problem of the allocation of resources for these sorted chains of functions. A new distributed algorithm, split in two distinct part, is introduced. The first part computes a path that takes into account the constraints on the order of the resources and their placement on the network topology. The second part allocates the resources using vectors of counters for the scheduling or requests and a preemption mechanism to enforce it. Several heuristics are proposed for both parts. A numerical method is proposed to compare the performance of the algorithm to the expected value. The performances are then compared with those of four algorithms from the state of the art on an evaluation platform based on the SimGrid simulator. Results shows an improvement of up to 20% of the Average Usage Rate while not degrading the other metrics.