Network Controller Bandwidth Management for Non-Blocking Spine Leaf Fabric
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional spine and leaf network architectures in data centers face challenges in managing bandwidth for long-lasting large flows, such as live video production, which require a non-blocking, zero-loss fabric with low latency and minimal jitter, while also dealing with faults and exceeding single spine capabilities, leading to underutilization and reachability issues.
Innovation Solution
A network controller discovers topology and link capacities, selects optimal spine nodes and uplinks for flow transmission, and implements self-healing mechanisms to ensure non-blocking behavior across multiple paths, providing detailed visibility and minimizing underutilization by dynamically managing uplinks and downlinks between spine and leaf nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional spine and leaf network architectures are used, then the network can handle typical data center flows, but the network experiences underutilization and reachability issues when dealing with long-lasting large flows such as live video production
Solution Approach 1:
The patent implements dynamic flow migration capabilities that allow flows to be moved between different spine nodes based on real-time network conditions, receiver availability, and load balancing requirements. This dynamic adaptation enables the network to optimize bandwidth utilization while maintaining reachability for long-lasting large flows in live video production scenarios.
Solution Approach 2:
The system dynamically changes routing parameters and path selections based on network state. The controller monitors link capacities, node loads, and receiver positions, then adjusts flow paths accordingly. This parameter adaptation resolves the contradiction by enabling both high bandwidth utilization and reliable reachability through intelligent, condition-based routing decisions.
2Productivity
If multiple spine nodes are used to exceed single spine capabilities, then the network can handle larger flows, but the complexity of managing bandwidth across multiple paths increases
Solution Approach 1:
The patent introduces a centralized controller as an intermediary that manages bandwidth allocation and flow routing across multiple spine nodes. The controller receives flow setup requests, determines optimal paths based on current network state, and programs the switching fabric accordingly. This intermediary abstraction simplifies bandwidth management complexity while enabling the network to exceed single spine capabilities through coordinated multi-spine operation.
3Productivity
If the network prioritizes non-blocking behavior for video production, then bandwidth utilization improves, but the network becomes more sensitive to faults and requires self-healing mechanisms
Solution Approach 1:
The patent implements monitoring and self-healing mechanisms that detect faults and trigger flow migration before complete service degradation occurs. The system maintains awareness of receiver positions and network conditions, allowing it to proactively relocate flows around failures. This beforehand cushioning enables the network to maintain high bandwidth utilization for video production while becoming resilient to faults through automated recovery processes.
4Loss of time
If flows are transmitted proactively before receiver requests, then latency is reduced, but bandwidth may be wasted if no receivers are requesting the flow
Solution Approach 1:
The patent implements a hybrid approach where the controller pre-computes and prepares optimal flow paths based on predicted receiver positions and network conditions, but actual flow transmission is triggered only when receiver requests are detected. This preliminary preparation of routing information reduces setup latency when flows are needed, while the conditional transmission trigger prevents bandwidth waste by ensuring flows are only activated when receivers are actually requesting them.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In one embodiment, a method includes discovering at a network controller, a topology and link capacities for a network, the network controller in communication with a plurality of spine nodes and leaf nodes, the link capacities comprising capacities for links between the spine nodes and the leaf nodes, identifying at the network controller, a flow received from a source at one of the leaf nodes, selecting at the network controller, one of the spine nodes to receive the flow from the leaf node based, at least in part, on the link capacities, and programming the network to transmit the flow from the spine node to one of the leaf nodes in communication with a receiver requesting the flow. An apparatus and logic are also disclosed herein.