Distributing Web-Scale Network Attributes: Hyperscale for the Masses
October 1, 2022

Distributing Web-Scale Network Attributes: Hyperscale for the Masses

Enterprises today face chaotic situations akin to those experienced by Internet web-scale businesses, also known as hyperscalers, in the year 2010 when bursty and distributed traffic exploded and choked their data infrastructure. These situations are brought on by the Internet of Things (IoT), a variety of data stores with structured and unstructured data, and X-Reality devices.

To keep up with the exponential surge in traffic, the hyperscalers—among them Amazon, Google, and Facebook—had to drastically scale their data centres. The internet giants realised that throwing hardware at the issue was necessary but insufficient to provide the solution they required, so they came up with a plan to scale-out their data centres by automating their operations with technologies like software-defined networking (SDN) and network virtualization, which allowed them to quickly adapt to the characteristics of the diversity of traffic with the resources at hand.

Business Data Centers for the Digital Revolution

In order to meet the needs for flexibility, variety, scalability, and availability, enterprise data centres have not only virtualized their computing but are also seeking to obtain software agility across storage, compute, and networking.

In order to respond to huge event-related data arriving from sources like security cameras at the edge, applications like “AI on the Fly” need recalibrating artificial intelligence (AI) algorithms, parsing of live data, and its display.

Collocation sites’ data architecture allows for quicker data input and analysis without putting a strain on the network infrastructure.

Of course, hyperscale architecture for the business and the edge will be different from internet hyperscale; it will be more likely to imitate their characteristics than their scale.

“The principles and innovations that Internet Hyperscalers brought to their data centres can be adapted and repackaged (downsized and simplified) for enterprise customers and their application environments as they have evolved with digital transformation,” said Brad Casemore, Research Vice President, Datacenter Networking, IDC.”

“Businesses may create networks with hyperscale characteristics like as resilience, availability, and dependability via superior architectural designs and cloud operating models, as well as agility through intelligent automation, flexibility, and programmability.

Additionally, in order to boost performance, computing power is being distributed out from centralised company data centres to locations closer to consumers and devices. This spreads to collocation facilities and then as far as what is referred to as the edge. Certain use cases and workloads, such as those at the edge, which are restricted by the costs, latency, and privacy of moving traffic to the public cloud, will benefit most from enterprise hyperscale.

According to Mike Capuano, CMO of Pluribus Networks, “privacy is crucial for use cases such as face recognition at a theme park since consumers don’t want the data to leave their jurisdiction.”

Analytics is arguably quicker and less expensive at the edge for concentrated data streams, as those from security cameras at sporting venues.

Hyperscale Networks for the Edge that resemble clouds

The characteristics of hyperscale, according to a growing collection of networking businesses, including Big Switch, may be achieved at considerably smaller sizes. Kyle Forster, the inventor of Big Switch, said during a presentation at the NetEvents conference in San Jose that open switches already have extra processing power that can be exploited by software-defined networks and made available to all businesses, including smaller ones.

The phrase “hyperscale for the people” was coined due to the widespread use of switches.

Web-scale networks were designed and built by Internet hyperscale enterprises using a massive skill pool of engineers, which the sector has found difficult to replicate on its own. “We have developed a complete networking automation package that the enterprise, particularly mid and small-sized organisations, can utilise without having a big pool of technical knowledge. It is all pre-integrated with SDN, virtualization, network analytics, and network segmentation. Data centre networks may be automated by even the smallest IT teams, according to Capuano.

Co-founder, CDO, and chairman of Arista Andreas Bechtolsheim said that clients are modernising campus networks and enterprise data centres using cloud concepts to enhance performance, resilience, flexibility, security, and management. He does not see many new corporate data centres being constructed, either, since 15% of all US enterprise data centres shuttered their doors in 2017 and it is expected that another 35% will do so by 2022. Consolidation had a role in this, but switching to the public cloud is the main cause.

Hyperscale at the edge does not, in Bechtolsheim’s opinion, have a future. Customers will instead be drawn to metro clouds, which run at significantly cheaper rates for the same value, to host their apps. “Packets can make a round journey over a distance of 100 kilometres in only one millisecond, which makes cloud computing more efficient than servers that are situated at the edge. The economics of larger cloud data centres are intrinsically superior than scattering servers along the perimeter, according to Bechtolsheim.

“Economics always prevails in the end.” Emerging applications, like the Internet of Things, are not restricted to a single area but rather are dispersed over various locations, which the cloud can better service, he said.

A few high-density applications, like video in sports stadiums or robots in factories, may provide prospects for edge hyperscale, but Bechtolsheim claims that there is presently no commercial justification for them. Bechtolsheim said that “sports stadiums sometimes create massive amounts of data for a short period of time which is not enough to build a hyperscale network at the edge.”

Regarding robotics, he spoke of a recent visit he had to a Volkswagen factory where the whole wiring network for the facility—which houses thousands of robots—had been created internally to guarantee stability and zero downtime.

The remaining aspects of applications will, in his opinion, be handled by C-RAN, 5G, and fibre optic networks connecting to a public cloud at affordable prices.

In order to attain the needed levels of service quality, enterprise data networks are obtaining hyperscale cloud-like qualities like agility and intelligence. One such is the Hyper-Scalable Enterprise Framework from Mellanox, which employs network interface cards (SmartNICs) and I/O Processing Units (IPUs) with integrated processors for making decisions in real-time at the edge.

Kevin Deierling, VP of Marketing at Mellanox, said that “intent-based networks fulfil pre-determined objectives for service quality by, for example, automatically identifying and nullifying cyberattacks, and skirting around malfunctioning systems to reduce downtime.”

The NVIDIA EGX Edge Supercomputing Platform, unveiled in 2019 at Mobile World Congress, will benefit from Mellanox’s recent acquisition by NVIDIA, which will provide network performance solutions through its adapters and switches. According to Deierling, “Nvidia’s platform can analyse fast flowing data from factory floors, manufacturing inspection lines, and city streets with no latency.”

With SONiC (Software for Open Networking in the Cloud), a spin-off from Microsoft’s Azure, an open network operating system for ethernet switches, elasticity comparable to that achieved by cloud computing will be realised. SONiC disaggregates software from hardware and reduces it into microservices that customers can use to customise their services.

“In a leaf-and-spine network design, enterprise hyperscale may scale progressively employing virtualized IT resources. As the need for applications like AI and ML grows, they may be incorporated into systems like OpenStack or Kubernetes to manage storage, computing, and networks, according to Capuano.

Final Reflections

New entrants now have new potential to create networks that are specialised for client demands because to an explosion of data at the edge. Over the last ten years, centralised cloud computing has become a very successful business model. Its success is supported by lower costs and enticing value.

There are enticing use cases for the edge, and the new competitors are putting business models in place that cater to their particular requirements.

Without turning to cheap sources of power, the new entrants rely on various strategies to keep costs down while still adding value in order to stay competitive. These strategies include open networking and the spare capacity of switches.

Leave a Reply

Your email address will not be published. Required fields are marked *