Packet Switching Device Load Balancing via Policing Drop Decisions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In packet switching networks, traditional congestion management methods often result in packet dropping, which can lead to suboptimal performance and increased latency due to the inability to effectively manage traffic congestion without disrupting ongoing TCP sessions.

Innovation Solution

Implementing a policing discipline that determines congested packet flows and applies load balancing by reassigning output queues or paths, using methods like random reassignment or Deficit Weighted Round Robin, to avoid packet dropping and maintain TCP session performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional congestion management methods are used to drop packets during congestion, then packet loss is reduced, but TCP session performance deteriorates and latency increases

Engineering Contradiction:
Improvepacket loss preventionVSAvoidlatency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent introduces an intermediary mechanism (ECN marking and load balancing redirection) between the congested network condition and the TCP session. Instead of directly dropping packets which terminates the flow, the system uses ECN marks as an intermediary signal to indicate congestion, triggering load balancing redirection that preserves the TCP session while缓解 congestion. This intermediary approach prevents the direct harm of packet dropping on TCP performance.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the parameter of packet handling from binary drop/forward to a three-state system (forward, ECN mark, redirect). By introducing ECN marking as an intermediate state and using load balancing redirection as a dynamic parameter change (changing the output queue assignment), the system maintains packet flow while responding to congestion. This parameter expansion allows the system to preserve TCP sessions while managing congestion effectively.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If packets are dropped during congestion to manage traffic, then congestion control is achieved, but network throughput decreases

Engineering Contradiction:
Improvenetwork throughputVSAvoidcongestion control
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies dynamics by making the output queue assignment dynamic rather than static. When congestion is detected via ECN marks, the load balancing mechanism dynamically redirects flows to different output queues, adapting the network path in real-time. This dynamic reassignment maintains throughput by keeping packets moving while providing responsive congestion control, avoiding the static limitations of traditional drop-based methods.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent implements feedback through ECN marking, where congested network conditions are signaled back to the sending end and intermediate devices. This feedback mechanism allows the system to respond to congestion without blind packet dropping, enabling informed load balancing decisions that maintain throughput while achieving congestion control. The feedback loop prevents unnecessary packet loss and maintains network productivity.

Inventive Principle:
Principle #23Feedback

3Productivity

If load balancing is applied to all packet flows, then network utilization improves, but device complexity increases

Engineering Contradiction:
Improvenetwork utilizationVSAvoidload balancing complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent applies local quality by implementing load balancing selectively rather than uniformly across all packet flows. ECN-marked packets trigger load balancing redirection, while normal packets follow their original path. This localized application of load balancing optimizes network utilization for congested flows without unnecessarily complicating the handling of non-congested traffic, reducing overall device complexity while maintaining productivity benefits.

Inventive Principle:
Principle #3Local quality

4Reliability

If TCP sessions are allowed to enter slow start state during congestion, then congestion recovery is achieved, but performance degradation occurs

Engineering Contradiction:
Improvecongestion recoveryVSAvoidsession performance
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by proactively redirecting ECN-marked packets to alternative output queues before the TCP session can enter the slow start state. This preemptive load balancing redirection prevents the performance degradation associated with slow start while still achieving congestion recovery. By acting beforehand rather than waiting for the session to degrade and then recover, the system maintains both reliability and productivity.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3334101B1Load balancing eligible packets in response to a policing drop decision
Publication Date: 2020.12.02 CISCO TECHNOLOGY INC
  • EP3334101B1 patent drawingFigure 1
  • EP3334101B1 patent drawingFigure 2
  • EP3334101B1 patent drawingFigure 3A~3B

AI summary

Embodiments include a packet switching device load balancing eligible packets in response to a policing drop decision. The packet switching device sends packets of a particular packet flow out of the packet switching device over a first path in the network towards a destination node; and in response to a policer discipline determining to drop a particular packet of the particular packet flow, switching from said sending packets over the first path to sending packets of the particular packet flow out of the packet switching device over a second path in the network towards the destination node (possibly by switching output queues associated with the two different paths), with the second path being different than the first path, and with the particular packet not being dropped but being sent out of the packet switching device towards the destination node.