Article Preview
Top1. Introduction
Intent-Based Networking System (IBNS) integrates with machine learning to ease and automate some of the administrative works for network administrators and engineers. IBNS enables the conventional practices that require individual network-element configurations to be replaced by an abstraction layer that easily enables operators to express intent and subsequently validate the network to do what they desired to perform on the network. Unlike traditional tab completions found in some modern command line interface(cli); IBNS understand intentions context and can provide tailored solution recommendations when deployed at solution architectural level (Han, Li, Hoang, Yoo, & Hong, 2016). This simplifies the dev/ops of network solution architecting by improving implementation decisions, agility and fostering security with advanced automation.
Solution architecting pipeline starts with high level policy making; usually from managerial people to solution architects to design a holistic network topology and finally to system administrators or network engineers for implementational works (Charalambides, et al., 2005). It is clear this is a complex pipeline that involves people with diverse IT literacy who are dealing with different constraints. Network policies are penned at a high abstraction level to account for business goals that benefits the organisation globally; which sometimes do not consider technical challenges and implementation feasibility to align with cost and performance constraints (Lupu & Sloman, 1999). The translation of intent to policies is a loosly process; whereby written policies in some cases do not reflect the original intents; especially when the gap of technical proficiency is profound. On technical fronts, another layer of loosly process happens when practitioners work to architect robust solutions to deploy the goals stated in the layman styled policies. This is an optimisation problem, where the objective function is to achieve high level goals using services, tools and technology that are attainable, scalable and manageable that are also at the same time, fitting to existing infrastructure. Optimal solutions are often case specific thus not easily carbon copied from some best practices architecture (Siau, Nah, & Teng, 2002). At the backend, there are ambiguous methods to measure optimality; whether it is latency, throughput or availability. At this stage, the experience of technical personnel in handling multi vendor solutions like ensuring interoperability between across Huawei, Cisco, 3com, Ericsson equipments might determine the robustness of resulting architecture. In addition, deep optimisation like conflict detection or synergy check among possible solutions often triumph deploying out of box solutions (Marcon, Dischinger, & Gummadi, 2011).
These constraints are recently addressed with Policy-based Network Management (PBNM). PBNM can be integrated with SDN, to delegate the entire management can encapsulate within a central unit that is smart enough to understand intents in policies and recommend contextual aware solutions (Avolio, Fallin, & Pinzon, 2007). In PBNM, the atomic unit is called policy. Policy, defined as the combination of rules and services where rules defined the criteria for resource access and usage. Each of the policies are composed a set of conditions and corresponding rules to overcome it. If the conditions of the policy rule are met, one or more actions contained by that policy rule may be executed. (Ding 2010). Policy-based management has simplified the complex task of managing networks and distributed systems. It could reduce the complexity of managing a large computer system and the manager does not need to constantly monitor the equipment and systems. PBNM allows admin to define various kinds of traffic metric such as data, voice, and video and assigns their priority of availability and bandwidth based on policy statements. The framework for PBNM over SDN use Neural Networks to monitor network parameters will adaptively react upon the policy violation. The use of the OpenFlow protocol enables the controller to communicate with the network devices to implement these changes. Meanwhile, VivoNet (Chaudhari, et al., 2019) was developed by Django that use voice assistance to capture networking intents. Leveraging on NLP, the components of VIVoNet are runs as VMs on VMWARE ESXi hypervisor are trained to accept high level voice input for configuration jobs. The novelty comes from ease of configuration; but ViVoNet works on some pre-trained intents instead of some general purpose intents that are normally found on written network policies.