Cloud Tales – Comics
AI Networking Installation
InfiniBand and Ethernet are both powerful networking technologies with unique strengths in different contexts. While InfiniBand traditionally excelled in high-performance computing environments, Ethernet AI Networking Fabric has evolved to meet the demands of modern data centers and AI applications.
Network Convergence
It may seem like a contradiction, but the disaggregation of network elements facilitates the convergence of both network hardware and operations.
Ethernet
Fabric-Scheduled Ethernet guarantees lossless connectivity for a large-scale server array running high-bandwidth workloads free of flow discrimination and with minimal impact from network failures, all in a traditional and easy-to- manage Clos-like topology.
Job Completion Time
As the volume and complexity of AI models continue to grow, reducing job completion time (JCT) and improving related back-end networking efficiency have become critical objectives.
Open Relationships
The AI/ML market is rapidly growing, driving greater diversity of GPUs and AI ASICs. The network will play a key role in enabling the new processors to enter the market. Ethernet is adapting to support these changes in the most open and flexible ways possible.
Job Completion Time
As the volume and complexity of AI models continue to grow, reducing job completion time (JCT) and improving related back-end networking efficiency have become critical objectives.
Pizza Box
Recently, we announced support for white boxes based on Broadcom’s Qumran 2C+. This extends our solution to more edge and aggregation use cases and demonstrates Network Cloud’s flexibility in supporting a choice ASICs and white boxes.
Until the Network Edge
Learn how to build a modern edge network – software-based over white boxes.
Lost in Barcelona
We’re happy to share that DIGI deployed DriveNets in core sites across Romania, demonstrating the smooth transition from traditional networks to the cloud-like disaggregated architecture. Also, we made a big splash at MWC in Barcelona.
Simple Operations
Building a network like cloud brings benefits not only in terms of innovation and savings, but also in terms of operations from the planning and installation and throughout the operations the infrastructure.
AI Workloads
AI workloads demand high-speed data transfers, low-latency communication, and efficient resource utilization. In AI networks, any failure or delayed flow can hold back the progress of all nodes.
InfiniBand vs Ethernet
To align with the ever-growing demands of data centers, selecting a networking technology for hyperscale data centers is more crucial than ever. For data transmission, InfiniBand and Ethernet, compete for supremacy.
Halloween in the Data Center
As large cloud providers have adopted data center server applications and associated spine & leaf deployment (3-stage Clos), service providers have also considered Clos architecture to scale out large core or aggregation routers, even though their applications may not entirely align with the traffic pattern and applications in the cloud provider’s data center. Service providers must carefully evaluate whether a traditional Clos solution satisfies carrier-class performance, unique traffic flow patterns with a multi-service router, and the cost of its complex distributed control and data plane features.
AI Networking
How can a Distributed Disaggregated Chassis (DDC) for AI Networking Fabric, already proven as a robust and scalable solution in the telecom space, also make an optimal performance AI network a feasible reality? Learn more about AI Networking.
Disaggregated Network Infrastructure
Network disaggregation has enabled service providers to move toward more agile networks, offering much greater agility in addressing evolving requirements.
High Availability on Date Night
When building a network infrastructure for mission-critical traffic, high availability is an uncompromising requirement.
Network Edge: Until the end of the world
As more and more latency-sensitive applications emerge – networking and compute functions need to move towards the network edge – closer to the end user.
Network White box Cluster
DriveNets introduced a “disaggregated” networking model of a cluster of white boxes and separate software.
Peering at the Bar
Peering allows two networks to connect and exchange traffic directly, without having to pay a third party to carry traffic across the Internet.
Network Scaling
We live in a scaling world, driven by the hunger for more capacity, and for more services and functionality.
Minimize Network Downtime
Even a brief amount of downtime can lead to lost revenue, damaged reputation, legal liability, and brand dilution from a business perspective..
Stayin’ Alive with the Border Gateway Protocol
BGP, or the Border Gateway Protocol, is the gateway protocol that enables exchanging routing information between autonomous systems (AS), through peering. BGP is really a kind mail service for the Internet.