A Comparison of Markov Reward Based Resource-Latency Aware Heuristics for the Virtual Network Embedding Problem

The increasing use of virtualization (e.g., in Cloud Computing, Software Defined Networks), demands Infrastructure Providers (InPs) to optimize the placement of the virtual network requests (VNRs) into a substrate network. In addition to that, they need to cope with QoS, in particular for the rising number of time critical applications (e.g., healthcare, VoIP). Granting resource optimization along with QoS compliance are two competing goals. In this work, we compare our QoS-aware virtual network embedding (VNE) algorithms. A first approach (MCRR-LA) relies on a powerful resource-latency aware metric, whilst a second one (MCRM) building on the former one adds a node proximity and similarity concept, with a common goal of achieving an effective ranking/mapping. We extend the MCRR-LA metric and the MCRM node/link mapping stage for testing new performing strategies. We widely evaluated our algorithms through simulation. Our experiments point out that our algorithms are able to reduce the average path delay while granting good resource performances in terms of lower VNR blocking rate and higher revenues. We compared our algorithm with a previous two-stage approach obtaining good results useful to underline the strengths of the novel approaches.