The Rising Star: AI Networking Steals the Spotlight at ONUG Summit

This week, the Open Networking User Group (ONUG) focused on the realm of AI networking.

As the utilization of AI expands, the foundation that supports both current and future possibilities lies in the strength of the network. This sentiment resonated throughout the ONUG AI Networking Summit, which took place in New York and was available via webcast.

Preparing the network for AI presents both technical challenges and process hurdles. AI transcends being merely another type of data traffic; it is also a technological advancement capable of enhancing network operations, a significant theme addressed during the event.

“We have entered the engineered miracle economy afforded by AI to improve the human condition,” remarked Nick Lippis, co-founder and co-chairman of ONUG, during the opening keynote. “We are infrastructure professionals; this is our golden age. We have the opportunity to enhance human experience by constructing and empowering these miracles to take shape.”

The demand for hardware and bandwidth driven by AI presents significant challenges for connectivity across Wide Area Networks (WAN).

During a panel discussion addressing the transition from conventional WAN to AI-enhanced solutions, Rajarshi Purkayastha, who serves as the vice president of customer strategy and presales at Tata Communications, highlighted that GPUs and AI workloads require exceptionally high bandwidth, frequently reaching hundreds of terabits per second. This level of demand makes traditional WAN connections impractical and economically unviable.

Purkayastha stressed the crucial necessity for new standards and reference architectures that would facilitate the incorporation of GPUs into various devices, spanning from smartphones to Internet of Things (IoT) gadgets.

“We are going to see GPUs integrated into everyday devices, including phones, laptops, and IoT devices, which indicates that the existing network framework will need to undergo a significant transformation to meet the future demands posed by GPUs,” he stated.

Artificial Intelligence is set to significantly enhance the speed of deploying and managing WAN capabilities.

Ilwyn Sequeira, the founder and CEO of Highway 9 Networks, discussed how their company successfully leveraged AI and machine learning to drastically shorten the implementation and configuration timeframe for their private mobile cloud solution on a campus setting.

“We recently executed a complete installation at MIT, specifically in a vital eight-story building, which features multiple radios connecting with macro towers,” Sequeira explained. “Following the initial configuration phase, we applied AI/ML techniques for the day one setup, which reduced our implementation and configuration timeline from several weeks to just four or five hours.”

During the ONUG event, the use of AI for enhancing network automation was a prevalent topic. ONUG presented an AI-Driven NOC/SOC Automation Project that outlined its developments during a panel discussion.

The group has recently finalized a study concerning AI automation within the NOC that has not yet been made public. In a session, several key findings from the report were disclosed. One significant application of generative AI in the NOC is utilizing chatbots to assist users. When inquired about the main advantages of generative AI in this context, the most common answer was its potential to enhance the productivity and effectiveness of operations teams.

The insights from the survey were supported by the panelists, who shared some practical examples. Parantap Lahiri, vice president of network and data center engineering at eBay, mentioned that his organization is currently implementing AI for a networking monitoring system. This system is capable of processing and analyzing a substantial amount of log and alert data, enabling humans to prioritize their needs effectively.

Xiaobo Long, head of backbone network services at Citi, highlighted that her organization employs AI chatbots to alleviate resource limitations faced by the network team.

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

The future of networking is increasingly leaning toward autonomous, self-driving functionalities; however, the journey to achieve this is fraught with obstacles. During the ONUG session focusing on network configuration, participants engaged in discussions around the progression of network automation alongside the integration of AI.

While automation has been a part of networking for some time, the notable advancement with AI lies in transitioning from pure automation to an augmentation model.

Mark Berly, Chief Technology Officer for data center networking at Aruba, a subsidiary of Hewlett Packard Enterprise, remarked that innovations like zero touch provisioning have broadened the scope of automation in recent years. He emphasized that automation is well-established for known and existing processes. The real advantage of AI is its ability to augment these processes. The key difference is that prior automation primarily focused on defining specific use cases and workflows.

In contrast, the panelists view AI-enhanced network automation as a movement toward more flexible, autonomous functionalities that can address unforeseen circumstances, rather than merely executing predetermined tasks.

While the dream of entirely automated networks is on the horizon, achieving this reality will require time and effort. Berly shared his experience with a self-driving car, humorously mentioning that it has put him in danger at least once, leading him to limit its use to just parking.

“We are definitely moving closer to that fully autonomous phase, and frankly, it does concern me. My car nearly caused me an accident, and I can’t help but think that my network might try to crash itself too,” Berly remarked.

With the rapid rise of generative AI (GenAI), the strain on current network capacities and structures is becoming increasingly important to address.

During a panel discussion, Gerald de Grace, a cloud architect and technical product manager at Microsoft, underscored the vast scale of the challenge, pointing out that the company is managing clusters featuring more than 300,000 GPUs. The massive volume of components involved makes failures unavoidable, prompting de Grace to stress the need for automated systems capable of swiftly identifying, isolating, and resolving these problems.

“Our goal is to integrate autonomous systems and AI to quickly identify and eliminate failing components from the network,” he stated.

The panel also discussed the implications of InfiniBand versus Ethernet.

De Grace mentioned that Microsoft is currently amenable to using InfiniBand, as it is the standard for the GPU stack. However, he pointed out some operational challenges that InfiniBand poses for the company. He noted that Microsoft is observing initial solutions based on Ethernet that support RoCEv2 (RDMA over Converged Ethernet), and he anticipates these will be tested in smaller data centers in the coming year or two. Following that, he believes Microsoft will likely shift from InfiniBand to more Ethernet-based networking for AI workloads.

The primary reasons for this transition, according to Grace, are the operational ease and cost benefits of Ethernet compared to InfiniBand. He emphasized that training all their engineers on InfiniBand is “just another burden,” and the company prefers to concentrate on Ethernet-centric solutions that can be seamlessly integrated into their current infrastructure.

Citi’s Long highlighted the importance of standardizing protocols and interfaces for AI networking, regardless of whether it involves InfiniBand, GPU connectivity, or other technical aspects.

“We consistently focus on standardization, striving for simplified technology across various environments. This approach is always the best practice, whether or not we are looking to support AI,” she stated.

Total
0
Shares
Leave a Reply

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

Previous Article

Chinese Hackers Exploit Verizon Breach to Target Trump Campaign: What You Need to Know

Next Article

Unmissable Deal: Resident Evil 4 Remake for PS5 Now Just $20 on Amazon!

Related Posts