The New AI Era: Networking for AI and AI for Networking*

The New AI Era: Networking for AI and AI for Networking*

As we all recover from NVIDIA’s exhilarating GTC 2024 in San Jose last week, AI state-of-the-art news seems fast and furious. Nvidia’s latest Blackwell GPU announcement and Meta’s blog validating Ethernet for their pair of clusters with 24,000 GPUs to train on their Llama 3 large language model (LLM) made the headlines. Networking has come a long way, accelerating pervasive compute, storage, and AI workloads for the next era of AI. Our large customers across every market segment, as well as the cloud and AI titans, recognize the rapid improvements in productivity and unprecedented insights and knowledge that AI enables. At the heart of many of these AI clusters is the flagship Arista 7800R AI spine.

NB471: Nvidia Unveils 800G Ethernet, InfiniBand Switches For AI Fabrics; ‘Ghost Jobs’ Haunt Job Boards

Take a Network Break! Nvidia announces new 800G switches, one for Ethernet and one for InfiniBand, for building AI fabrics. Nvidia also announces an “AI supercomputer,” a rack-scale pre-built bundle of Nvidia GPUs and CPUs connected via InfiniBand switches. The NaaS startup Meter announces new campus switches and what it calls a “digital twin” capability,... Read more »

netlab: Building Leaf-and-Spine Fabrics with the Fabric Plugin

netlab release 1.7.0 added the fabric plugin that simplifies building lab topologies with leaf-and-spine fabrics. All you have to do to build a full-blown leaf-and-spine fabric is:

  • Specify the default device type
  • Enable the fabric plugin
  • Specify the number of leaves and spines in the fabric.

For example, the following lab topology builds a fabric with Arista cEOS containers having two spines and four leaves:

Read more …

netlab: Building Leaf-and-Spine Fabrics with the Fabric Plugin

netlab release 1.7.0 added the fabric plugin that simplifies building lab topologies with leaf-and-spine fabrics. All you have to do to build a full-blown leaf-and-spine fabric is:

  • Specify the default device type
  • Enable the fabric plugin
  • Specify the number of leaves and spines in the fabric.

For example, the following lab topology builds a fabric with Arista cEOS containers having two spines and four leaves:

Read more …

Azure Networking: Cloud Scale Load Balancing

 Introduction


During the load balancer deployment process, we define a virtual IP (a.k.a front-end IP) for our published service. As a next step, we create a backend (BE) pool to which we attach Virtual Machines using either their associated vNIC or Direct IP (DIP). Then, we bind the VIP to BE using an Inbound rule. Besides, in this phase, we create and associate health probes with inbound rules for monitoring VM's service availability. If VMs in the backend pool also initiate outbound connections, we build an outbound policy, which states the source Network Address Translation (SNAT) rule (DIP, src port > VIP, src port).  

This chapter provides an overview of the components of the Azure load balancer service: Centralized SDN Controller, Virtual Load balancer pools, and Host Agents. In this chapter, we discuss control plane and data plane operation.  


Management & Control Plane – External Connections

Figure 20-1 depicts our example diagram. The top-most box, Loadbalancer deployment, shows our LB settings. We intend to forward HTTP traffic from the Internet to VIP 1.2.3.4 to either DIP 10.0.0.4 (vm-beetle) or DIP 10.0.0.5 (vm-bailey). The health probe associated with Continue reading

Weekend Reads 032224


Highway 9 Networks recently emerged from stealth mode with $25 million in seed funding and a vision to provide enterprise companies with SaaS-based private mobile networks that would eliminate gaps in coverage by incorporating cellular technology.


Hackers’ advantage: One of the biggest security weaknesses in U.S. digital networks and infrastructure is out-of-date, no-longer-supported technology.


Threat actors have been observed leveraging the QEMU open-source hardware emulator as tunneling software during a cyber attack targeting an unnamed “large company” to connect to their infrastructure.


There are some weighty ironies here. The AI “safety” experts had raised alarm about “implicit bias” in AI, only for Google to release an almost parodically racist chatbot.


Yet another DNS vulnerability has been exposed. The language of the press release revealing the vulnerability is certainly dramatic, with “devasting consequences” and the threat to “completely disable large parts of the worldwide Internet.”


It’s a good rule of nostril that if your litigation department is a source of revenue, your business model stinks.


In his January 12 SpaceX update, Elon Musk said the biggest goal for Starlink from a technical standpoint is to get the mean latency below 20 ms.


In a recent press release, John Deere Continue reading

MMLU in Network Flow Optimization

In a previous post, I discussed how Maximum Flow problems can be used for network optimization. We focused on a scenario where demands were already routed in the network, and our objective was to determine the maximum demand that could be handled between a given source and a destination metro. We solved this problem by calculating the residual bandwidth for the graph, creating fake demand nodes for each metro with high-capacity edges to avoid them being bottlenecks, and applying Dinic’s algorithm between the source and the destination metro. This is also called a Single Commodity Flow Problem.

MMLU in Network Flow Optimization

In a previous post, I discussed how Maximum Flow problems can be used for network optimization. We focused on a scenario where demands were already routed in the network, and our objective was to determine the maximum demand that could be handled between a given source and a destination metro. We solved this problem by calculating the residual bandwidth for the graph, creating fake demand nodes for each metro with high-capacity edges to avoid them being bottlenecks, and applying Dinic’s algorithm between the source and the destination metro. This is also called a Single Commodity Flow Problem.

We then extended the problem to consider two metros sending traffic to the same destination sink and used the Network Simplex algorithm to determine the maximum traffic the network could accommodate. This is also known as a Multi Commodity Flow Problem. Finally, we validated our findings by routing the results through a network model.

In this post, we will discuss another constraint-based problem called Minimum Maximum Link Utilization (MMLU). The primary goal of MMLU is to route traffic demands in a network to minimize the maximum link utilization. In other words, we aim to distribute the traffic evenly across the network links to Continue reading

HW023: The Best of WLPC 2024 Phoenix

The Wireless LAN Professionals organization just had its 10th annual conference and who better to break it down than WLPC founder (and Heavy Wireless host) Keith Parsons and friend of the show Ferney Munoz. They review their favorite presentations as well as heartwarming moments. Episode Guest Ferney Munoz | Ekahau and CWNP Certified Wireless Network... Read more »
1 … 147 148 149 150 151 … 3,834