This week, the Open Networking User Group (ONUG) turned its attention to the realm of AI networking.

As the adoption of AI continues to rise, the essential element that supports both current and future capabilities is the strength of the network. This sentiment was reiterated multiple times throughout the ONUG AI Networking Summit, which took place in New York and was available via webcast.

Facilitating the network for AI presents both technical and procedural hurdles. AI represents more than just an additional type of data traffic; it is also a technology that has the potential to enhance network operations, which was another significant topic discussed during the event.

“We have entered the engineered miracle economy afforded by AI to improve the human condition,” stated Nick Lippis, cofounder and co-chairman of ONUG, during his opening keynote. “As infrastructure professionals, we are in our golden age; we have the opportunity to enhance the human experience by building and enabling these extraordinary advancements.”

Given the substantial hardware and bandwidth demands of AI, there are significant challenges related to AI connectivity over the WAN.

During a panel discussion focused on the transition from WAN to AI-optimized solutions, Rajarshi Purkayastha, who serves as the vice president of customer strategy and presales at Tata Communications, pointed out that GPUs and associated AI workloads require exceptionally high bandwidth, frequently reaching hundreds of terabits per second. Connecting these workloads via a traditional WAN is neither practical nor economically viable.

Purkayastha highlighted the necessity for the establishment of new standards and reference architectures that will facilitate the integration of GPUs into a diverse array of devices, including phones and IoT technologies.

“We will see GPUs in a multitude of end-user devices, such as phones, laptops, and IoT technologies, which signifies that the current network infrastructure will need to undergo a substantial transformation to meet the future demands of GPUs,” he remarked.

Artificial Intelligence is poised to enhance the speed of deploying and provisioning Wide Area Network (WAN) capabilities.

Llwyn Sequeira, the founder and CEO of Highway 9 Networks, elaborated on how his company has harnessed AI and machine learning to dramatically cut down the time required for implementation and configuration of their private mobile cloud solution at a campus.

“We recently completed a full build at MIT, specifically in one of the primary eight-story buildings, which involved several radios connecting with macro towers,” Sequeira explained. “Following the initial configuration phase, we employed AI and machine learning techniques for the day one setup, reducing the overall implementation and configuration time from weeks to just four or five hours.”

The discussion on leveraging AI to enhance network automation emerged as a central theme at the ONUG event. Notably, ONUG is advancing an AI-Driven NOC/SOC Automation Project, which showcased its developments during a panel session.

The team has recently finalized a study concerning AI automation within the NOC, which has yet to be disclosed to the public. During a session, several significant findings from the report were shared. A prominent application of generative AI in the NOC is the implementation of chatbots to assist users. When inquiring about the main advantages of generative AI in the NOC, the leading answer indicated that it could enhance the productivity and efficiency of operations teams.

The insights from the survey were reaffirmed by the panelists through various real-world examples. Parantap Lahiri, who serves as vice president of network and data center engineering at eBay, mentioned that his company is currently leveraging AI for a network monitoring system. This system can sift through and evaluate a substantial amount of log and alert data, aiding humans in prioritizing their needs.

Xiaobo Long, who leads backbone network services at Citi, remarked that her organization employs AI chatbots to tackle resource limitations faced by the network team.

“I am confident that chatbots will save us a significant amount of time, allowing our team to concentrate on resolving more intricate issues for our customers,” she stated.

The future of networking is steering towards autonomous, self-driving functionalities, but the journey to achieve this vision presents several hurdles. The ONUG session dedicated to network configuration engaged in dialogues about the advancement of network automation and the integration of AI.

While automation has long been a component of networking, the notable change with AI lies in the progression from simple automation to augmentation.

Mark Berly, the Chief Technology Officer of data center networking at Aruba, a division of Hewlett Packard Enterprise, highlighted that innovations like zero touch provisioning have broadened the scope of automation in recent times. He pointed out that for familiar and established processes, automation is already firmly in place. The advancement with AI is its ability to enhance and supplement existing methods. The key difference lies in that earlier forms of automation primarily focused on delineating specific use cases and workflows.

In contrast, the panelists perceive AI-driven network automation as a transition towards more flexible, autonomous capabilities equipped to manage unforeseen challenges, rather than being limited to just preset tasks.

The journey towards fully automated networks, where operations are entirely handled by machines, may be on the horizon, but patience is required. Berly shared his experience with a self-driving car, humorously mentioning its near-miss that led him to rely on its abilities solely for parking purposes.

“I genuinely believe we are edging closer to that autonomous phase, and honestly, it terrifies me. My vehicle nearly caused an accident, and I can only imagine my network attempting to fail on its own,” Berly expressed.

As generative AI (GenAI) becomes more widely embraced, the effects on present network capabilities and designs are raising significant concerns.

During a panel discussion, Gerald de Grace, a cloud architect and technical product manager at Microsoft, addressed the monumental scale of the challenge, pointing out that the organization is examining clusters that incorporate over 300,000 GPUs. The vast array of components involved suggests that failures will occur, prompting de Grace to highlight the necessity for automated systems capable of swiftly detecting, isolating, and addressing these problems.

“To effectively address the issues, we need to integrate autonomous systems and AI technologies that can identify failing components, isolate them, and promptly remove them from the network,” he stated.

The discussion also covered the comparison between InfiniBand and Ethernet.

De Grace indicated that while Microsoft currently accepts InfiniBand, as it is the standard for the GPU framework, there are operational challenges associated with it. He mentioned that the company is noticing promising initial developments in Ethernet-based solutions that support RoCEv2 (RDMA over Converged Ethernet), and he anticipates these technologies will be validated in smaller data centers within the next couple of years. Following this period, he believes Microsoft will likely shift from InfiniBand to more Ethernet-focused networking for AI applications.

According to Grace, the primary factors driving this transition include the operational ease and cost-efficiency of Ethernet in comparison to InfiniBand. He pointed out that training their engineers on InfiniBand adds to their workload, and the organization prefers to concentrate on Ethernet solutions that can seamlessly fit into their existing systems.

Regardless of whether it involves InfiniBand, GPU connectivity, or various other technical aspects of AI networking at scale, Citi’s Long stressed the importance of establishing standardized protocols and interfaces for AI networking.

“Continuous standardization is essential; we should always strive for simplified technology across our diverse environments. This approach remains a best practice, whether we are focused on supporting AI or not,” she stated.

Welcome to DediRock, your trusted partner in high-performance hosting solutions. At DediRock, we specialize in providing dedicated servers, VPS hosting, and cloud services tailored to meet the unique needs of businesses and individuals alike. Our mission is to deliver reliable, scalable, and secure hosting solutions that empower our clients to achieve their digital goals. With a commitment to exceptional customer support, cutting-edge technology, and robust infrastructure, DediRock stands out as a leader in the hosting industry. Join us and experience the difference that dedicated service and unwavering reliability can make for your online presence. Launch our website.