Clustering algorithms for dynamic adaptation of service function chains

Network function virtualization is a pillar-stone of today’s network architectures as it offers better management and elasticity and allows also a flexible maintenance of services running on shared resources over cloud environments. Network functions traditionally hosted on dedicated hardware are now provided over software based components that might run either on virtual machines or on containers. The major advantage of this transition is that it makes the deployment of new services easier while optimizing the management and administration of network architectures. It is much easier to spin up a new virtual machine/container hosting a network function or a specific application described as a service function chain, than to deploy a new hardware based equipment and checking its compatibility with the rest of the architecture. With all the advantages that this new paradigm offers comes a set of challenges related mainly to: 1) optimizing the resource consumption on the shared infrastructure 2) making the best decision of placing the virtual functions that respects at the same time clients’ requirements and also leverages the available resources on the substrate network in terms of different metrics (e.g., CPU, memory, latency, bandwidth). This aspect of Network Function Virtualization-NFV and Service Function Chains-SFC placement have been treated in so many research works that propose approaches ensuring optimal placement and chaining of VNFs in virtualized networks, but as the adoption of these technologies gets more important in real network setups, and given the strict restrictions of today’s’ applications (e.g. latency highly-sensitive applications, or availability highly-sensitive service, etc.), it is always important to consider all the parameters impacting the network management in cloud environments. In this research project, we develop new approaches for placement and chaining of virtual network functions in cloud-based environments. The first approach allows forming on demand clusters of servers deployed in a physical infrastructure. These servers are grouped according to their similar attributes (e.g., CPU-intensive server, energy-efficient server, etc). This process is a proactive measure to ensure that SFCs are hosted in servers that meet their specific metrics requirements (CPU, memory, disk, etc.). It employs a meta-heuristic called CRO (Chemical Reaction Optimization) to decide of the best VNF placement guaranteeing optimal resource consumption in terms of CPU / memory. We employ CRO also to ensure the lowest latencies during the routing between the different VNFs. In fact, the E2E delay is an important aspect to consider, as most current applications require low latencies and shortest run times. In the second approach, the clusters are formed using algorithms based on meta-heuristics, including the CRO, allowing to improve the quality of clusters formed in terms of similarity, density and modularity