Upstarts harvest digital services on heterogeneous virtual networks
Infusing creativity in networks for a long-tail of services
By Kishore Jethanandani
A third generation of competitors are transforming software-defined networks; they harvest digital services from a rich repository of resources on heterogeneous networks.
The amassing of IT resources and network management services on software-defined networks, in a single pool as it were, accelerates a pattern of innovation spawned by virtualization. Upstarts, who provide innovative digital services, have ready access to the resources that can feed agile software development supported by Docker containers. By separating application development components from operating systems, Docker containers compound the benefits of virtualization or software that is divisible from hardware.
Stating intent on interfaces
Interfaces are the means the upstarts use to reach distributed network assets and to orchestrate the assembly of components. On these interfaces, “applications state their intent, or their need to fulfill a need, relying on complementary components in a network reference model to have the resources to achieve them. On the network side, the reference model corresponds to functional blocks, such as storage, computing power, databases, switching, and orchestration, which have their north-south or east-west relationships. Beyond stating their intent, developers are freed from the technical and the execution complexity,” Dan Pitt, Senior Vice-President of the MEF Forum told us. “The individual layers are separable and combined on-demand, automatically, with intent-based networking without being encumbered by the details within each layer,” Dan Pitt added.
The first generation of software-defined networks were the giant Internet companies, such as Google and Facebook, who took advantage of commodity hardware to lower costs that they interconnected with software to meet their internal needs for scalable and flexible networks for digital services.
Web-scale networks serve two different purposes; the first meets the internal needs of a few applications of Internet giants while the second has opened to a wider market for a broad range of customer applications. Lately, Amazon’s AWS has been a pioneer in opening the cloud and other services for enterprises and start-ups. “Since the primary motivation is to advance the operator’s cloud business, they are not designed to benefit from the emerging world of multiple clouds, virtualized radio networks, and software-defined service providers and have not yet been optimized to serve the growing demands of the Internet of Things, the global enterprise, streaming media, or autonomous cars,” Dan Pitt told us.
Today’s customers are best served by the use of a diversity of resources. The global enterprise, for example, communicates across the world with multiple clouds, streaming media needs content delivery networks, and IoT uses radio networks to communicate with devices.
Traditional telecom service providers have responded to the competitive pressures from web-scale networks and virtualized their networks, in fits and starts, to cater to the spiky demand from digital services. They sought to take advantage of their proximity to customers that their Ethernet networks afford. However, service providers are constrained to using the resources available on their networks. Even when they have the full range of assets within their networks, they are not necessarily able to balance loads by spreading them over several networks.
The third generation of networks
The third generation of competitors in software-defined networks doesn’t necessarily own any of the networks. Instead, they lease the resources they need, from a diversity of networks, and interconnect them virtually. They focus their energy on efficiently managing traffic with modern technologies like artificial intelligence, intent-based networking, and portals for command and control.
PCCW Global, a telecom service provider with headquarters in Hong Kong, acquired Console Connect, a software-defined automation platform and integrated it with their Tier 1 global IP network to provide point and click private connections to a host of SaaS applications, cloud service providers and multi-national enterprises. “SaaS providers want to extend their geographic footprint and their addressable market which can be challenging for them because hybrid and multi-cloud environments are hard to manage,” Neil Templeton, VP Strategic Marketing at PCCW Global, told us.
Digital services grow with ease on a platform, with capabilities to develop applications, such as the one provided by Console Connect. “We have nTwine, our globally hosted UCaaS service built around a Softswitch platform which uses voice switching and web collaboration capabilities complete with OSS/BSS and management capabilities,” Neil Templeton informed us. Such services are white-labeled for regional service providers who want to provide their own branded digital applications incurring only operating costs. Often these regional players don’t have ready access to cloud providers but do when they connect with PCCW Global. “Wider adoption of cloud-based white label services by regional service providers will happen with multi-tenancy connectivity that serves multiple customers,” Neil Templeton added.
Webscale networks customized
Networks of the future encounter multiple challenges in managing the bandwidth, the costs, and the quality of experience as their heterogeneity grows. The potential solutions are often unique to the situation of each user. Incumbent cloud providers are either unable to tailor solutions for their customers or their offerings are much too expensive.
Netrolix, one of the third generation of competitors, created its own network fabric by interconnecting countless resources on the Internet, across all types of networks, with its sixty-eight data centers and twenty thousand nodes spread around the world which provides opportunities for load balancing. Customers can connect with the Netrolix network fabric from an on-site MEC endpoint (edge computing) equipped with their proprietary software and routing engine which costs much less than the offerings from incumbent cloud providers. Its development platform helps to craft applications customized for the service needs of its customers. “The IoT is fraught with security issues which customers can combat by segmenting their network with role-based connections which are monitored with help from AI-aided cybersecurity,” Wes Jensen, the CEO, and co-founder of Netrolix told us.
Additionally, the network fabric lays the groundwork for the control of traffic for individual customers. “It provides visibility into every Internet network on the globe as well as the access and endpoints which enables us to gather network performance data,” Wes Jensen said. “Our AI algorithm finds the most optimal routes for moving traffic for our customers,” he explained.
Optimizing software-defined networks
A need for new ways to route exploding traffic has grown to make the best of alternative network routes tailored for the demands of individual types of data flows. “CPUs, the most expensive element in networks, are best utilized for the optimal navigation of workloads while the hardware layer is more efficient for moving large volumes of data with ASICs,” Kevin Deierling, Vice-President of Marketing for Mellanox Technologies, surmised. “In recent years, virtual switches which bring the advantages of programmability have played a growing role in the transportation of data, as virtual machines on networks have increased in numbers, which has led to a sharp deterioration in performance,” Kevin Deierling, observed.
Heterogenous networks, furthermore, encounter choke points at intersections where traffic moves from one type of network to another. “For video traffic, CDNs have been widely deployed for efficiently moving traffic to the edge or a co-location facility. The distribution of content at rates proportionate to the capacities of set-top boxes, in millions of homes, is done more efficiently with ASICs,” Kevin Deierling informed us. “Moving traffic from one cloud to another is inefficient because it navigates multiple domains and management structures. At the hardware layer, the traffic moves as if on a single cloud,” he explained. “Similarly, traffic on C-RANs crosses multiple hoops as millions of calls, received at towers, are directed to servers for decoding before they are sent to the intended receivers. All the steering of traffic is done from a single point at the hardware layer,” Kevin Deierling concluded.
Automating network management
The fluidity in the demand for digital services, the type, scale, and quality of service, creates operational challenges in the configuration and validation of networks that will deliver them — repeatable methods are needed to respond quickly to market changes. “Apstra’s operating system has reference designs and blueprints (topology), invoked from a user interface, that recognize a use case or a service and automatically generate the corresponding network configurations and network state data to be used for validation. New reference designs are created for emerging applications,” Manish Sampat, VP of Engineering at Apstra told us.
Apstra’s operating system responds to rare events by reading them from telemetry data. “Special mathematical routines are created for certain events to facilitate the detection of anomalies. Telemetry provides the environmental data and its correlation with the network state to achieve the desired intent,” Manish Sampat explained. Apstra’s operating system allows customers to have the option to interact with the graph database and programmatically reason about intent and build respond to deviations from intent. It is possible to leverage technologies like AI to reason and build intent-based responses.
The third generation of competitors in software-defined networks has infused creativity in the control and orchestration of networks. Unlike the OSS and BSS systems of yore, network management systems are continually evolving as they learn from data and find new ways to optimize traffic management and adapt to customer needs. They lower the barriers to entry for new applications. The networks of the future will support a long-tail of services.