Leaf Switch Uplink Bandwidth Advertisement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In spine-and-leaf datacenter network architectures, the need for inter-leaf switch connections to manage network traffic imbalances leads to resource inefficiency, as these connections consume port bandwidth that could be used for uplink capacity, especially as datacenters scale, resulting in costly resource allocation.
Innovation Solution
The solution involves removing the requirement for adjacent physical connections between leaf network switches, shifting the responsibility for traffic distribution from leaf switches to servers, which advertise their uplink bandwidth to each spine network switch, allowing servers to distribute network traffic based on available bandwidth, thereby optimizing port usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If inter-leaf switch connections are established to manage network traffic imbalances, then network traffic distribution is improved, but port bandwidth is consumed that could be used for uplink capacity
Solution Approach 1:
The patent extracts the inter-leaf switch connections from the network architecture entirely. Instead of having leaf switches physically connected to each other for traffic redistribution, the solution removes these connections and shifts the traffic distribution function to the servers themselves through ECMP routing, thereby freeing up port bandwidth for uplink capacity.
Solution Approach 2:
The patent implements self-service by enabling servers to autonomously perform load distribution across multiple leaf switches using ECMP (Equal-Cost Multi-Path) routing. Each server independently selects which leaf switch to send traffic to based on available paths, eliminating the need for centralized inter-leaf coordination and manual traffic engineering.
2Reliability
If leaf switches are physically interconnected to provide load balancing, then traffic distribution is improved, but device complexity increases
Solution Approach 1:
The patent removes the complex inter-leaf physical connections and the associated leaf switch intelligence required for traffic redistribution. By extracting this functionality and placing it at the server level through standard ECMP routing, the network architecture becomes simpler with fewer physical connections and less switch complexity.
Solution Approach 2:
Instead of having leaf switches perform load balancing functions (which requires complex switch logic and inter-leaf connections), the patent inverts the approach by having servers perform the load balancing through ECMP routing. This shifts the intelligence from the network infrastructure to the endpoint devices, simplifying the network core.
3Reliability
If inter-leaf connections are reserved to ensure bandwidth resiliency, then network reliability is improved, but resource allocation efficiency deteriorates
Solution Approach 1:
The patent enables dynamic, on-demand traffic distribution through ECMP routing at the server level. Instead of pre-reserving bandwidth through static inter-leaf connections, servers automatically select optimal paths based on current network conditions, achieving resiliency without wasting resources on guaranteed but unused capacity.
Solution Approach 2:
The patent transitions from static bandwidth reservation through fixed inter-leaf connections to dynamic traffic distribution where servers adaptively select paths based on real-time conditions. This dynamic approach allows resources to be allocated flexibly according to actual demand rather than being locked into predetermined reservations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system includes spine network switching devices, leaf network switching devices, and server computing devices. The leaf network switching devices are not connected to one another. Each leaf network switching device is connected to each spine network switching device. Each server computing device is connected to each leaf network switching device. Each leaf network switching device transmits an advertisement indicating uplink bandwidth to each spine network device. Each server computing device distributes network traffic through the leaf network switching devices to the spine network switching devices based on the uplink bandwidth that the leaf network switching devices advertise.