When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
The enterprise in the age of heterogeneous data flowing from the Internet of Things (IoT), a diversity of data stores with structured and unstructured data, and X-Reality devices encounters the chaotic situations similar to those that Internet web-scale companies, also known as hyperscalers, did around 2010 when bursty and distributed traffic exploded and choked their data infrastructure.
The hyperscalers, such as Amazon, Google, Facebook among others, had to massively scale their data centers to cope with the exponential growth in traffic. Throwing hardware at the problem was necessary but not sufficient for the solution the internet giants needed; instead, they improvised to scale-out their data centers bringing to bear software to automate their processes with technologies like software-defined networking (SDN) and network virtualization to flexibly adapt to the attributes of the diversity of traffic with available resources.
Enterprise Data Centers for Digital Transformation
Enterprise data centers have not only virtualized their compute but are also attempting to acquire software agility, across storage, compute, and networking, to cope with the demands for adaptability, versatility, scalability and availability.
Applications such as “AI on the Fly” require recalibration of artificial intelligence (AI) algorithms, parsing of live data, and its visualization, in response to massive event-related data coming from sources such as surveillance cameras at the edge. (Read Edge Computing: The Next Phase of IT.)
Data infrastructure at collocation sites can do the ingestion of the data and its analysis faster without taxing the network infrastructure.
To be sure, hyperscale architecture for the enterprise and the edge will be distinct from internet hyperscale—it will be inclined to emulate the attributes of Internet hyperscale companies, not necessarily the scale.
Brad Casemore, Research Vice President, Datacenter Networking, IDC, said: “The principles and innovations that Internet Hyperscalers brought to their data centers can be adapted and repackaged (downsized and simplified) for enterprise customers and their application environments as they have evolved with digital transformation"
"The enterprises can build networks with hyperscale attributes–agility through intelligent automation, flexibility through programmability, availability, and reliability through pervasive visibility, and resilience through better architectural designs and cloud operating models.”
Furthermore, computing power is getting dispersed from centralized enterprise data centers to sites closer to users and devices to improve performance—they spread to collocation facilities and then as far as what is described as the edge. Enterprise hyperscale will best support certain use cases and workloads, including those at the edge, which are constrained by costs, latency, and privacy to move traffic to the public cloud.
“For use cases such as facial recognition in a theme park, privacy is paramount as customers don’t want the data to leave their precincts,” said Mike Capuano, CMO, Pluribus Networks.
“Concentrated data streams, such as those from surveillance cameras in sports stadiums, analytics is arguably faster and cheaper at the edge.”
Cloud-Like Hyperscale Networks for the Edge
An emerging group of networking companies, Big Switch, among others, believe that the attributes of hyperscale are achievable at much lower scales. Speaking at the NetEvents conference in San Jose, Kyle Forster (founder of Big Switch), explained that open switches already have the surplus processing power, which can be used by software-defined networks and made consumable by every enterprise including smaller ones.
The ubiquity of switches has sparked the slogan “hyperscale for the masses.”
Internet hyperscale companies drew on a vast pool of engineering talent to design and construct web-scale networks that the enterprise, on its own, has found hard to emulate. “We have created a full networking automation package that is all pre-integrated with SDN, virtualization, network analytics, and network segmentation which the enterprise, especially mid and small-sized companies can use without needing a deep pool of engineering talent. The smallest IT teams can automate their data center networks,” said Capuano.
Andreas Bechtolsheim, co-founder, CDO, and chairman of Arista, agreed that customers are upgrading Enterprise data centers and campus networks with cloud principles to improve performance, resilience, flexibility, security and manageability. However, he does not see a lot of new enterprise data centers being built; 15% of all enterprise data centers in the US closed last year, and the prediction is by 2022 another 35% will close. Some of this is due to consolidation, but the primary reason is the move to the public cloud.
Bechtolsheim does not believe that hyperscale at the edge has a future. Instead, customers will be attracted to metro clouds, for hosting their applications, which operate at much lower costs for the same value. “Cloud computing performs better than servers located at the edge because packets can complete a round trip in a 100-kilometer range in one millisecond. Larger-scale cloud data centers offer inherently better economics than placing a bunch of servers around the edges,” said Bechtolsheim. (Read Cloud Computing and Cloud Servers: How Do You Know Your Cloud Data is Protected?)
“In the end, economics always wins.” “Emerging applications such as the Internet of Things are not confined to one location but are distributed over several geographies which the cloud can serve better,” he added.
A few high-density applications, such as video in sports stadiums or robotics in factories, have the potential to be opportunities for edge hyperscale but the business case for them, according to Bechtolsheim, does not exist currently. “Sports stadiums generate huge volumes of data, on occasion, for a short period which is not enough to have a hyperscale network at the edge,” Bechtolsheim said.
As for robotics, he recounted his visit to Volkswagen recently where an entire wired network for the plant, with thousands of robots, has been built in-house to ensure reliability for zero downtime. (Read How does AI interact with robotics?)
For the rest, he believes C-RAN, 5G, and fiber optic network to connect with a public cloud will take care of all the desired attributes of applications and provide connectivity at economical rates.
Enterprise data networks are acquiring hyperscale cloud-like characteristics such as agility and intelligence to self-remedy performance bottlenecks, in real-time, to achieve the desired levels of service quality. An example is Mellanox’s Hyper-Scalable Enterprise Framework, which uses SmartNICs (Network Interface Cards) and I/O Processing Units (IPUs) with built-in processors for decision-making, in real-time, at the edge.
“Intent-based networks realize pre-determined goals for service quality by, for example, automatically recognizing and nullifying cyberattacks, and skirting around broken systems to avoid downtime,” said Kevin Deierling, VP of Marketing, Mellanox.
Mellanox, recently acquired by NVIDIA, will complement the NVIDIA EGX Edge Supercomputing Platform, announced in 2019 at the Mobile World Congress, with network performance solutions with its adapters and switches. “Nvidia’s platform will be able to process rapidly streaming data from factory floors, manufacturing inspection lines, and city streets with minimal delay,” said Deierling.
Elasticity, comparable to that achieved by cloud computing, will be realized with SONiC (Software for Open Networking in the Cloud), spun-off from Microsoft’s Azure, an open network operating system for ethernet switches, which disaggregates software from hardware and reduces it into microservices that customers can use to customize their services.
“Enterprise hyperscale can scale incrementally, using virtualized IT resources, in a leaf-and-spine network architecture. These, in turn, can be integrated into systems such as Openstack or Kubernetes, that can orchestrate storage, computer, and networks for applications like AI/ML as their demand expands,” Capuano explained.
Final Thoughts
An explosion of data at the edge has opened new opportunities for new entrants to design networks tailored for customer needs. Centralized cloud computing is a proven business model that has soared over the last decade. Lower costs and compelling value are the underpinnings of its success.
There are tantalizing use cases for the edge; the new entrants are implementing business models that address their unique needs.
The new entrants are counting on alternative ways to keep costs low such as open networking and the spare capacity of switches, without recourse to low-cost sources of power, and create value to remain competitive.
