Our teams at NWN Carousel recently attended Cisco Live, where a series of groundbreaking announcements were made, set to revolutionize the future of networking and data centers. The standout highlight was the unveiling of HyperFabric, an innovative networking architecture designed to meet the unique demands of AI systems. HyperFabric is more than just a technological advancement, it necessitates substantial changes in the design and operation of colocation facilities and data centers.
AI Focus – has always been here.
Cisco emphasized that it has been incorporating AI and machine learning technologies into its product portfolio for many years. What’s truly new is generative AI (GenAI) and the big impact it’s having on the tech industry. This trend also applies to NWN Carousel’s core offerings. AI is integrated throughout our portfolio, with expertise in cloud transformation, intelligent infrastructure, cybersecurity, managed devices, visual collaboration, and cloud communications delivering AI-powered technology solutions for the modern workplace.
HyperFabric: The New Network Paradigm
HyperFabric is set to revolutionize network infrastructure, catering to the needs of AI systems that demand networks far beyond traditional data or storage networks. Historically, high-performance networks for AI were built on custom InfiniBand setups. The industry now is transitioning towards more cost-effective Ethernet networks, which offer both scalability and affordability.
Lifecycle Considerations
For typical customers, understanding the lifecycle of network equipment vs. server equipment is crucial. In most organizations, servers are replaced two to three times during the lifespan of the network infrastructure. Traditionally, network advancements have outpaced server bandwidth requirements. However, the rapid advancements in AI workloads are changing this dynamic, akin to the shift from 1 GbE to 10 GbE during the virtualization boom.
Data Sovereignty and Cloud Integration
While some AI models will inevitably reside in the cloud, many will require substantial training data or leverage proprietary data with stringent data sovereignty issues. This means that colocations and data centers must be equipped to handle both cloud-integrated and on-premise AI workloads efficiently.
Adapting to Elephant Flows and Simultaneous Demand
The simultaneous demand across the fabric and the prevalence of elephant flows (large, persistent data transfers) necessitate a hyper fabric that is distinct from traditional networks. We are already witnessing the segregation of data, storage, and AI fabrics within data centers. Selecting the appropriate equipment is vital to avoid bottlenecks that could leave expensive AI servers underutilized.
Power and Cooling: The New Norms
The introduction of AI workloads is also reshaping fundamental aspects of data center infrastructure, such as power and cooling requirements. Unlike current norms where 3-phase power is rare, the new AI workloads will necessitate high power per rack and enhanced cooling solutions. Although current equipment still relies on air cooling, the next generation is expected to transition towards more efficient cooling methods. Organizations must prepare for racks consuming as much power as an entire row did just a few years ago.
Enhancing Observability with AI
Observability is evolving with the rise of AI augmenting traditional machine learning systems. This trend is liberating resources and funding to collect the data necessary for both new AI systems and existing machine learning models. As the industry grapples with a shortage of skilled personnel, these advanced tools are essential for automating routine tasks and allowing experts to focus on complex problem-solving.
A Shift in Security
On the security front, we anticipate a fundamental shift, though traditional methods will be hard to abandon. The advent of AI pods and dedicated fabrics for GPU workloads demands a departure from the “god box” firewall mentality. The pace of this transformation will vary across organizations, influenced by industry, compliance, and emerging insurance requirements. A cross-functional, architectural overhaul in data center security is expected to be challenging and time-consuming, likely outpacing initial expectations.
Digital Experience Assurance
Cisco ThousandEyes enables organizations to detect, diagnose, remediate, predict, and optimize conditions affecting connected experiences across any domain. It introduces enhancements, such as deeper integration of ThousandEyes agents into Meraki and Catalyst products and embedding ThousandEyes data into the Meraki SD-WAN manager. It includes the Traffic Insights feature for linking traffic activity with user experience and embedding ThousandEyes experience measurements within Cisco’s Secure Access client for enhanced user experience and security. Following the recent acquisition of Splunk, Cisco is integrating Splunk into its strategic direction, focusing on security and linking it with AppDynamics and full-stack observability.
Transformative Period
The Cisco Live event underscored a transformative period for network and data center architecture. HyperFabric, with its AI-centric design, will redefine how colocations and data centers operate, pushing the boundaries of what current infrastructures can support. As organizations navigate these changes, strategic planning in lifecycle management, power and cooling adaptations, enhanced observability, and revamped security protocols will be essential to harness the full potential of AI and next-generation workloads.
NWN Carousel experts to guide you on your modern work journey,
Connect with Us