Dynamic and Scalable Control as a Foundation for Future Networks

Dynamic and Scalable Control as a Foundation for Future Networks

Zoran Despotovic (Huawei, Germany), Xun Xiao (Huawei, Germany), Ramin Khalili (Huawei, Germany), Maja Curic (Huawei, Germany) and Artur Hecker (Huawei, Germany)
Copyright: © 2019 |Pages: 23
DOI: 10.4018/978-1-5225-7146-9.ch008

Abstract

The authors see problems with current network control models. Their control networks (i.e., control channels, necessary for control operation) are not thought of as part of the control model itself. Current network control is not transactional. Network updates are neither atomic nor isolated, and the application is not aware of the details of an update outcome. This chapter presents an alternative design in which the control channel is an integral part of the network control model. Its key part is a robust, in-band resource connectivity layer that interconnects all available network elements, including the controller(s). The control is also transactional. Applications can safely assume that their updates will not clash in the network, as well as that they will always affect the right, intended fraction of the network. Building on these two postulates, the authors see service scheduling as its third essential part of network control. The scheduling takes service requirements into account and assigns the services network resources that will meet their requirements.
Chapter Preview
Top

Introduction

Network systems are becoming more and more customizable and programmable. This change has been exemplified through the development and the adoption of the three key technologies: Software-Defined Networking (SDN), Network Function Virtualization (NFV) and Service Function Chaining (SFC). NFV leverages virtualization technologies to implement Network Functions (NFs) on general purpose hardware platforms, in order to enable rapid creation, removal, or migration of NFs. SFC enables operation of composite services by steering traffic through an ordered set of network functions, thus providing network operators with the ability to more easily introduce new services and dynamically customize/change their runtime operation, e.g., to enforce various traffic forwarding policies for the sake of optimization. In that context, SDN emerges as an important automation and dynamicity enabler. As a technology that promotes programmability of transport network devices, it provides support for the said automation in the lower layers of the protocol stack.

Even though the three technologies exhibit important differences, often seeing the network from different angles (e.g., either merely using the network as is or rather interfering with the basic network primitives), the acquired levels of flexibility and efficiency, built on these three pillars, would represent a significant departure from the conventional networks, which are typically built upon closed hardware silos interconnected through overlays laid upon static, manually configured transport networks.

The authors believe that, taken together, SDN/NFV/SFC depict contours of the next network transformation: in this view, network becomes an execution environment (runtime) for requests of various kinds and requirements. In this chapter, the authors broadly refer to such requests as “jobs”. Such jobs could be, classically, to transport bits from point A to point B, possibly with transformations of the bits on the fly (e.g., header and address changes, filtering, compression), and, more generally, it could involve in-network processing of all bits and generation of replies. SFC, through both standardization efforts within the IETF (notably in the IETF SFC WG) and the ongoing academic research, sets up a model for what such jobs can be, and shows how the job execution progress is encoded and communicated in a chain of nodes that collaboratively execute it. In this light, SDN and NFV present the underlying network as a pool of resources that are to be instructed to execute individual parts of the entire job.

The upcoming 5G presents a perfect deployment environment for the just outlined vision. The three technologies, SDN/NFV/SFC, are current hot candidates for the realization of network slicing (An, et al., 2016), recently standardized by the 3GPP in their initial 5G-relevant Release 15 (TS23.501, 2017). Indeed, instead of simply defining a new particular system architecture, 5G bets on system flexibility. Release 15 supports finer-grained modularization, allows different flavors of standard-equivalent modules and opens up mobile systems of the future for the possibility to dynamically scale-in and scale out modules in operation. In particular, a network slice is a public land mobile network (PLMN) service, however composed not of standard physical entities, but instead of network functions that could be of different flavors and different realizations. In this way, the very same 5G infrastructure could be shared by several seemingly independent PLMNs corresponding to different business requirements. The proposed finer-grained modularization, the existence of the different flavors of network functions and their instantiation in different numbers of instances (scale-in and -out) ultimately all lead to an explosion of number of overall active modules in the system and the paths between them, every single one of which could play a critical role in the particular service provisioning.

In a simple view, NFV platforms could create, deploy and start/stop all the necessary 5G modules as virtual network functions (VNF), while specially crafted SDN control applications on an SDN controller could be used to route flows between those VNF instances. Note however that both the interactions between such individual VNFs and the performance of each VNF instance in the 5G scope should be better than best effort. Problem is that both recent NFV and SDN specifications and the popular NFV and SDN implementations do not directly support guarantees; in particular, they do not so far have any assurance mechanisms in their own realizations, raising doubts with respect to their suitability for 5G realizations, especially at the projected typical 5G scales.

Key Terms in this Chapter

R2R Protocol: A protocol used to create the controllability layer (i.e., autonomously establish interconnectivity among a set of network resources).

Self-Stabilizing Control Plane: A control plane always converges from an arbitrary status to a stable correct status in finite steps.

Transactional Network Updates: Network updates that execute as a whole and isolated from other updates.

State of Resources: Information about the set of active network entities, their service rates, and their traffic loads.

Job Request: An incoming network flow with a specific processing request.

Controllability Layer: A set of resources interconnected and configured to enable the controller-controlee (resource) communication.

Service Scheduling: Assignment of job requests to deployed, active network entities such that a preset goal is met.

Job Classification: The service function chain of a job; an ordered set of service functions that should be invoked by the job and the set of QoS requirements of the job.

Complete Chapter List

Search this Book:
Reset