Who’s watching your network bro?

Sriram Subramanian

Networking is going through a phase of rapid evolution with the advent of cloud computing. As a result, the world of network management has also seen some dramatic shifts.

Nothing is more representative of this change than the use of the word FCAPS. At one time the acronym FCAPS, Fault Configuration Accounting Provisioning and Security, was synonymous with network management (NM). But if you did a Google search for FCAPS now, you will find very few articles that use that term. Does it mean that we are not monitoring our networks anymore? How are we configuring our networks in the cloud computing age? Let us explore the drivers that are changing the way we operate our networks.

The rise of network virtualisation

As enterprise IT teams embrace cloud computing, applications and workloads are being moved to private clouds within the enterprise or to public clouds such as Amazon Web Service and Azure. This requires the cloud infrastructure to provide wide ranging compute, storage and network services. While server and storage virtualisation are relatively mature, the virtualisation of network is relatively a recent phenomenon.

Just like hypervisors (like VMware and KVM) virtualised server hardware, technologies like SDN are providing a software abstraction of the underlying physical network. Applications can then leverage these abstract or virtualised network services and become decoupled from the underlying physical network. The adoption of software-defined networks and network virtualisation has radically changed the way networks are managed. Let us now look at some key aspects of network management in the age of cloud computing.

Integrated and Automated

The main goal of the end users of cloud is to get their applications and workloads up and running with minimal effort. Applications components, storage and networking related requirements can be captured in templates and deployed in one click. This in turn means that network configuration happens in conjunction with application deployment and there is no direct-user intervention to configure networks.

Modern network-management applications must support programmatic interfaces (APIs) that allow orchestration tools like OpenStack, Kubernetes etc., to allow automatic configuration of virtual and physical networks. This integration can be enhanced by having automated topology discovery and path computation to allow orchestration tools to leverage the physical networks more efficiently while placing workloads.

Networks that heal themselves and others

Monitoring faults is one of the main capabilities of network management systems. Older NM applications displayed alarms and faults in a tabular view. The network operators used to monitor these views to take corrective actions. This is a time consuming, manual and error-prone mechanism and clearly not suited for the dynamic networking environments that support cloud computing.

Networking devices are now capable of streaming rich telemetry information that covers many aspects of the operation of the device. Network management systems can apply analytics and machine-learning techniques to the telemetry information and detect patterns and anomalies. This will form the basis for intelligent and self-healing networks that can detect faults and take corrective action. Additionally, the correction could be applied at the application level where workloads could be automatically moved away from parts of the network that are susceptible to faults.

Intent not Configuration

Traditionally operators and administrators have configured the networks using graphical management tools or using command-line interface. In a network with equipment from multiple vendors, this can be a challenge and requires additional tools. Even with that approach, the network configuration required deep knowledge of networking architecture and specialized teams to execute the changes needed to the network.

Recent developments indicate that the problem of vendor-agnostic network management can be solved using Intents. Instead of working at vendor-specific configuration, operators can specify their intent on how the network should respond to faults, to environmental change etc. This intent could be defined using natural language-based constructs or programmatic constructs. These constructs will then be translated into vendor-specific device configuration so that user operations are minimized in response to the changes detected in the networks.

Cloud-based network management

Finally, cloud computing has an even more direct impact on network management applications. Traditionally, network management applications were run using dedicated hardware appliance within the enterprise premises. The next change was to execute NM applications using virtual machine on hypervisors within the premises. The most recent trend is to host NM applications on public clouds. The trend has several implications on the design and architecture of network-management applications.

Even though the NM systems are moving to the cloud, the device being managed could still be within the enterprise premise protected by a firewall and NAT. This creates an interesting connectivity challenge for cloud-based network management applications. Furthermore, from traditional client-server architecture, NM applications on the cloud are migrating towards microservices-based architecture. This is required to leverage the auto-scaling and elasticity supported by cloud platforms.

Conclusion

As the cloud computing and software defined networking waves sweep the IT landscape; building reliable, efficient and automated networks is the key. Network management needs to evolve alongside to provide deep visibility and intelligent configuration for these modern networks.

(Sriram Subramanian is a technology professional with 20 years’ experience in building network management and cloud based applications; and has co-authored books titled 'OpenStack Networking Cookbook' and 'Software Defined Networking with OpenStack'. Sriram Subramanian is also Software Engineering Director at Juniper Networks IEC.
Views expressed here are personal)