Candidate: Junling Li
Title: Resource Allocation in SDN/NFV-EnabledCore Networks
Date: May 6, 2020
Time: 10:00 AM
Place: REMOTE PARTICIPATION
Supervisor(s): Shen, Sherman
Abstract:
For next generation core networks, it is anticipated to integrate communication, storage and computing resources into one programmable, flexible and unified infrastructure. Software-defined networking (SDN) and network function virtualization (NFV) become two enablers. SDN decouples the network control and forwarding functions, which facilitates network management and enables network programmability. NFV allows the network functions to be virtualized and placed on high capacity servers located anywhere in the network, not only on dedicated devices in current networks. Driven by SDN and NFV platforms, the future network architecture is expected to feature centralized network management, virtualized function chaining, reduced capital and operational costs, and enhanced service quality.
The combination of SDN and NFV provides a potential technical route to promote the future communication networks. However, it is imperative to efficiently manage, allocate and optimize the heterogeneous resources, including computing, storage, and communication resources, to the customized services to achieve better quality-of-service (QoS) provisioning. This thesis makes some in-depth researches on efficient resource allocation for SDN/NFV-enabled core networks in multiple aspects and dimensionality. Typically, the resource allocation task is implemented in three aspects. Given the traffic metrics, QoS requirements, and resource constraints of the substrate network, we first need to compose a virtual network function (VNF) chain to form a virtual network (VN) topology. Then, virtual resources allocated to each element (VNF or virtual link) need to be optimized in order to minimize the provisioning cost while satisfying the QoS requirements. Next, we need to embed the virtual network (i.e., VNF chain, or service function chain, SFC) onto the substrate network, in which we need to assign the physical resources in an economical way to meet the resource demands of virtual nodes and links. This involves determining the locations of NFV nodes to host the VNFs and the routing from source to destination. Finally, we need to schedule the VNFs for multiple services to minimize the service completion time and maximize the network performance.
In this thesis, we study resource allocation in SDN/NFV-enabled core networks from the aforementioned three aspects. First, we jointly study how to design the topology of a VN and embed the resultant VN onto a substrate network with the objective of minimizing the embedding cost while satisfying the QoS requirements. In VN topology design, optimizing the resource requirement for each virtual node and link is necessary. Without topology optimization, the resources assigned to the virtual network may be insufficient or redundant, leading to degraded service quality or increased embedding cost. The joint problem is formulated as an Mixed Integer Nonlinear Programming (MINLP), where queueing theory is utilized as the methodology to analyze the network delay and help to define the optimal set of physical resource requirements at network elements. Two algorithms are proposed to obtain the optimal/near-optimal solutions of the MINLP model.
Second, we propose a game theoretical approach to address the multi-VNF chain embedding problem with the impact of processing-resource sharing and node capacity constraints being considered. In the proposed game, each VNF chain is treated as one player whose objective is to minimize the overall latency experienced by its flow, while satisfying the capacity constraints of all the NFV nodes. The additional delay introduced by processing resource sharing is modeled and used in analyzing the overall latency for each VNF chain, while the node capacity constraints are handled by adding a penalty term to the cost function of each player. The proposed game is proved to be an exact potential game which admits at least one pure Nash Equilibrium (NE) and has the finite improvement property (FIP). We then design two iterative algorithms, i.e., the best response (BR) algorithm and the spatial adaptive play (SAP) algorithm, to find the NE of the game, which corresponds to a locally/globally optimal solution to the original problem.
Third, the VNF scheduling problem is investigated to minimize the makespan (i.e., overall completion time) of all services, while satisfying their different end-to-end (E2E) delay requirements. The problem is formulated as a mixed integer linear program (MILP) which is NP-hard with exponentially increasing computational complexity as the network size expands. To solve the MILP with high efficiency and accuracy, the original problem is reformulated as a Markov decision process (MDP) problem with variable action set. Then, a reinforcement learning (RL) algorithm is developed to learn the best scheduling policy by continuously interacting with the network environment. The proposed learning algorithm determines the variable action set at each decision-making state and accommodates different execution time of the actions. The reward function in the proposed algorithm is carefully designed to realize delay-aware VNF scheduling.
To sum up, it is of great importance to integrate SDN and NFV in the same network to accelerate the evolution toward software-enabled network services. We have studied VN topology design, multi-VNF chain embedding, and delay-aware VNF scheduling to achieve efficient resource allocation in different dimensions. The proposed approaches pave the way for exploiting network slicing to improve resource utilization and facilitate QoS-guaranteed service provisioning in SDN/NFV-enabled networks.