Resource provisioning for inter-domain IoT services in edge-cloud SDN-based networks

Abstract

The introduction of IoT networks with a large number of devices and diversity of smart IoT applications aims to improve the quality of human life. The heterogeneity of control and communication technologies from device to device, the fluctuating workload and low latency requirements of IoT applications pose a great pressure onto core network resource that traditional networks’ solutions are insufficient to handle while there are still obstacles that must be overcome before novel paradigms like SDN/NFV, edge-cloud computing paradigm can advance to be widely deployed. In particular, SDN controllers will face resource scalable challenges to deal with numerous queries from SDN forwarding elements for appropriate forwarding actions for traffic flows of IoT services that they encounter. On the other hand, in edge-cloud architecture, it is important to define a strategy to optimally allocate the virtual resources either at remote clouds for computation-intensive tasks or close to end-users at edge cloud on equipment within their premises to reduce E2E service latency. No prior work has taken into account the problem of virtual resource placement and chaining for edge-cloud SDN-based networks. Thus, the goal of this dissertation is to design efficient computing and networking resource placement and chaining strategy over both SDN control and forwarding planes. We present an E2E latency model for interworking services deployed over multiple domains that minimizes the total cost of resource usage and service operations while meeting QoS and QoE requirements as well as maintaining the system’s queue stability given dynamic service demand. In order to meet this goal, three key problems are to be addressed in our framework and are summarized as follows: i) how to optimize VNF resource placement and chaining in edge-cloud networks? ii) how to model and optimize resource placement and chaining solution in a heterogeneous network with a large number of nodes operated by SDN? iii) how to improve the QoS in terms of E2E service latency for multi-domain interworking services according to the availability of physical resources? Our first contribution is to model the VNF placement and chaining problem regarding aggregated input traffic from IoT devices. The E2E latency of service chains, the links connecting clouds, IoT gateways, and edges. To solve the formulated non-convex optimization problem, we design a Markov approximation-based solution that adopts multistart and batching techniques (MBMAP) to solve the combinatorial optimization problem. Our solution runs distributedly and consequently accelerates convergent rate. To address the second problem, we present a comprehensive model of resource allocation for both SDN controlling and forwarding planes and formulate it as a joint optimization problem. We adopt the Lyapunov queueing optimization framework to transform this long-term optimization problem into a series of real-time problems and employ the exponentiated gradient ascent method on the transformed problems to find a near-optimal solution. In addition, an implementation architecture for the orchestration of heterogeneous resource controllers is also designed. Finally, to address the third problem, we present an NFV-based interworking architecture enabling multi-domain orchestration, i.e. IMS and WebRTC domains, and provide the message exchanges required for service chains. Our E2E service latency model represents the real-time resource allocation objective. A real-time algorithm based on the Markov approximation framework is designed to allocate VNF resources with optimal cost and minimize impact from QoS violation during scaling periods. Experimental results reveal that our algorithm effectively responds to fluctuating service demands with a service cost reduced by 19% with respect to QoS requirements.

Date	3 Feb 2022
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Mohamed Cheriet (Supervisor) & Kim Khoa Nguyen (Co-supervisor)