While SD-WAN has moved out of its niche status and into the mainstream, it still needs added intelligence in order to see the big picture.
Global enterprise application performance has deteriorated as WAN traffic has grown exponentially in recent years driven by the expanded adoption of the cloud and the Internet of Things.
WAN traffic that traverses the public Internet has expanded as the traditional MPLS networks are costly to scale, slow and inflexible. Consequently, the variability of response times detrimental to application performance has increased. Investment in higher bandwidth to cope raises the risk of over-engineering and diminishing payoffs as the level of network utilization falls.
SD-WAN brings new capabilities for control and optimization of the flow of data traffic with centralized monitoring, SDN controller directed automated operations, distribution of traffic across the system to avoid local choke points, abstraction of complexity and reconfiguring the network to reduce the hops in data flows.
When the variability is out of range for network reconfiguration or optimization to work, private networks are constructed using existing nodes to gain greater network control. Above all, machine learning and artificial intelligence pinpoint the sources of inefficiency and possible remedies.
Increase in WAN traffic
A recent report by Aryaka revealed the variability in the performance of the public Internet on international routes. On a critical route between San Jose and Shanghai, the response time variation was 150% last year with an average of four seconds and a high of 40 seconds. On the Dubai-London route it was higher at 181%.
The poor response time is not surprising given the high rate of growth in the volume of WAN traffic, which was 200% year-over-year in 2016. There is no slowing down of the traffic growth as 50% of it in 2016 was accounted by HTTP/HTTPS that support cloud applications, according to the report. In manufacturing, WAN traffic zoomed from 296% in 2015 to 441% in 2017 as the adoption of IoT devices grew exponentially.
Enterprise customers have responded by increasing investments in higher bandwidth. The share of 100 megabits per second services grew from 7% in 2015 to 25% in 2016 (excluding direct datacenter connections) while 41Mbit/s to 99Mbit/s offerings increased from 9% to 17%.
Artificial intelligence for optimizing networks
UK-based Aria Networks Ltd. , which has used its platform of machine learning, genetic algorithms and other AI techniques to optimize physical and virtual networks for companies such Vodafone and Facebook, reported that the carriers' SD-WAN offerings use only a fraction of their existing capacity. Evidence of over-engineering to meet peak demand is conspicuous when seen from Aria's AI engines.
"The cornerstone of Aria's AI techniques is examining system-level performance, taking into account routing options, locations of assets, power consumption and more, for a collection of network nodes and links," said Jay Perrett, CTO and vice president of R&D at Aria Networks. "Aria’s platform optimizes for an objective function, such as faster response or lower costs, given an identified set of constraints, such as policy and energy costs and uncovers opportunities for improving network configurations."
Aria’s genetic algorithm-based technology can scale its objective function unsupervised to optimize networks with a larger number of nodes than Facebook could do on its own.
"A reduction of 25% in the number of IP ports between its [Facebook's] data centers was possible, by analyzing and optimizing its IP and optical network as a whole, rather than as separate layers," Perrett said.
That not only reduced the costs of moving traffic but also accelerates it as the number of hops are reduced.
Today’s SD-WANs provide a wide range of tools to streamline network configurations in order to improve performance. For example, they provide flexibility in the choice of pathways to achieve lower latencies.
"Historically application patterns have been client-server, for example, client to data center, so WANs had a hub-and-spoke topology," said Neil Anderson, practice manager of mobility and access solutions at World Wide Technology. "All the apps were in your data center. Now they are distributed in public cloud, SaaS platforms etcetera. Some apps need to go to a data center, but others more directly to where they are hosted, whether that's in Amazon Web Services Inc. or a SaaS provider like Salesforce.com Inc. or Microsoft Corp. (Nasdaq: MSFT) Office365.
"Other apps, like VoIP and collaboration, may need to flow more peer-to-peer to be successful and not be 'hair pinned' through the data center. Traditional WANs lacked that flexibility to specify application-dependent topologies."
The desired performance goals, such as faster and consistent response times, are not necessarily achieved by streamlining network configurations. Some companies have chosen to build private clouds or networks to reduce the variability in response times.
"Interoute uses a common SD-WAN, with an integrated core and edge, to provide an interconnected fabric, including software, network, storage, databases and computing, which enterprises can use for providing cloud services. We re-factor and abstract customers’ code to interconnect seamlessly," said Matthew Finnie, CTO at Interoute Communications Ltd.
Aryaka created a private network by leasing Layer 2 connectivity from Tier 1 networks globally with access to 95% of the Points of Presence(PoPs) used by business customers. With a private network providing consistent performance, Aryaka concentrates on the needs of the mobile and remote workforce.
"Our SmartAccess service, a clientless remote SD-WAN, allows users to access the nearest Aryaka PoP and enter its private network, which guarantees acceptable performance," said Gary Sevounts, chief marketing officer at Aryaka. "Application acceleration with dynamic CDN and WAN optimization software for the best routes and direct connection to clouds to avoid the Internet makes up for any performance degradation that might happen between users' devices and the PoP."
In order for SD-WAN to reach its full potential, sprawling deployments -- with heterogeneous resources, zigzagging routes, spikes in traffic and changing locations of its users -- need intelligence to see the whole picture in real-time, and SDN controllers to automate the operations.
— Kishore Jethanandani, Contributing Writer, Telco Transformation