For AI and IoT, Virtual Networks Can’t Do It All

One of the ideas behind virtual networking is that it allows organizations to create highly specialized architectures that suit the unique needs of individual applications, all in software.

While this may work for many of today’s back-office and customer-facing workloads, the fact remains that many emerging technologies require highly specialized environments of their own, and some of which require substantial changes to core infrastructure.

Take, for example, the increased prevalence of vision-based applications in the enterprise. These can range from facial recognition to augmented-reality (AR) infused video services, which not only stress existing networks due to their high-volume workloads, but must also be made available to numerous business functions such as training, security, collaboration and the like. A company called Userful Corp. recently released a new platform aimed at supporting visual services by creating a centrally managed architecture for all visual communications. The platform is available as a server-based enterprise deployment, as a cloud service, or as a display-based client under the LG webOS Signage platform. The result, according to the company, is an entirely new class of networking, dubbed a “visual networking platform,” that can improve operational efficiency by as much as 80 percent and reduce TCO by 40 percent. 

Artificial Intelligence (AI) is also throwing a curve ball at traditional network architectures, given its penchant for crunching massive volumes of data to influence the behavior of networks, applications and management software. Cisco’s AppDynamics subsidiary has put the rising field of AIOps in its sights with the new Central Nervous System monitoring platform that helps third-party systems ingest, correlate and analyze data across multiple domains as a means to troubleshoot problems and optimize performance. The system is based on three key components: a serverless agent for Amazon’s Lambda ecosystem, an application monitoring tool for the Cisco ACI platform, and a machine learning engine to glean operational insights into application environments.

Enterprises should also be aware that AI itself will also require changes to fundamental network infrastructure, says HPE’s Thomas Goepel. Most networks were built to support typical application workloads, mostly structured data, to on-site processing resources. AI workloads, on the other hand, support not only the parallel processing requirements of Big Data solutions like Hadoop, but must accommodate a world in which remote compute capacity is readily available on-demand and at scale. This means high-speed, always-available networking has become a critical element for organizations undergoing the transition to a digital services business model.

The Internet of Things presents another challenge, both for the enterprise and for regional network providers. As Chris Martin, CTO of conferencing software provider PowWowNow, explains, long-standing solutions like the Spanning Tree Protocol (STP) are in dire need of updating in order to produce adequate stability and reliability on Ethernet LANs. Meanwhile, alternate solutions like Multi System Link Aggregation (MLAG) and Shortest Path Bridging (SPB) are gaining support, leaving organizations with a dilemma as to whether they prefer higher link redundancy or loop-free multipathing and consolidation.

For the enterprise, perhaps the most difficult aspect of these changes is that they are being driven by application and user demands rather than advancements in technology. In the past, new networking capabilities were deployed, then, the user community set to work exploiting them to the greatest extent (and then immediately calling for newer, better technology). This time, the use cases are being defined first, and the networking team is expected to fulfill them or watch users gravitate to another network.

Making this happen on a virtual network is problematic enough; overhauling bare-metal infrastructure and basic communications protocols is going to take some time. And time is in increasingly short supply as the digital economy moves forward. 

Arthur Cole is a freelance journalist with more than 25 years’ experience covering Enterprise IT, telecommunications and other hi-tech industries. 

Latest Articles

Follow Us On Social Media

Explore More