Spine Switch Traffic Shaping for Leaf Buffer Overrun

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In data center networks with leaf-spine architecture, the insufficient buffer space at leaf layer switches often leads to packet drops due to buffer overrun when performing traffic shaping, as required by certain applications, since they lack the necessary capacity to handle data rate reductions from higher rates like 10 Gbps to lower rates like 1 Gbps.

Innovation Solution

The solution involves leveraging the larger buffer space in spine switches to perform traffic shaping, which are connected to leaf switches in a mesh, by instantiating traffic shapers on spine switches that buffer and reduce data rates for servers, thus avoiding buffer overruns at leaf switches.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traffic shaping is performed at the leaf layer switch, then traffic shaping function is achieved, but buffer space is insufficient causing packet drops

Engineering Contradiction:
Improvepacket delivery reliabilityVSAvoidbuffer space capacity
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent introduces the spine layer switch as an intermediary device between the external network and the leaf layer switch. The spine switch performs traffic shaping on incoming traffic before it reaches the leaf switch, using its larger buffer space capacity. This intermediary approach allows traffic shaping to be performed without overwhelming the limited buffer space at the leaf layer switch, thereby preventing packet drops while maintaining the traffic shaping function.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If data rate is reduced from 10 Gbps to 1 Gbps at the leaf layer switch, then traffic shaping is achieved, but buffer overrun occurs due to insufficient buffer space

Engineering Contradiction:
Improvedata rate adjustment capabilityVSAvoidbuffer space capacity
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The spine layer switch acts as an intermediary that performs the data rate adjustment (traffic shaping) from 10 Gbps to 1 Gbps before traffic reaches the leaf layer switch. The spine switch's larger buffer space accommodates the buffering requirements for this significant data rate reduction, preventing buffer overrun at the leaf switch while maintaining the required productivity of reducing data rate from 10 Gbps to 1 Gbps.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If leaf layer switch buffer space is increased to perform traffic shaping, then packet drops are prevented, but device complexity and cost increase

Engineering Contradiction:
Improvebuffer capacity for traffic shapingVSAvoidswitch buffer capacity requirements
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the traffic shaping function with the spine layer switch that already exists in the network architecture. Instead of increasing buffer space at the leaf layer switch, the solution combines the traffic shaping capability with the spine switch's existing larger buffer resources. This merging approach achieves the required buffer capacity for traffic shaping without increasing the complexity or cost of individual leaf layer switches.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS11196673B2Traffic shaping over multiple hops in a network
Publication Date: 2021.12.07 ARISTA NETWORKS INC
  • US11196673B2 patent drawing
  • US11196673B2 patent drawing
  • US11196673B2 patent drawing

AI summary

A first network device receives a message from a second network device with an indication that the first network device is to adjust a data rate of data being received by the first network device. The first network device includes a first buffer space that is greater than a second buffer space of the second network device. The first network device determines a set of computing devices connected to the second network device based on receiving the indication and defines a set of first data rates to use to send data to respective computing devices. The first network device adjusts a second data rate of received data to send the received data through the second network device to a computing device at a first data rate defined in the set of first data rates. The first network device uses the first buffer space to perform the adjusting.