NIC Clock Synchronization via Swarm Consensus

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current precision time protocol (PTP) technologies fail to provide tight NIC-to-NIC clock synchronization at scale and do not ensure synchronization with a universal time coordinate (UTC), especially in large-scale applications like financial systems.

Innovation Solution

The method involves network interface cards (NICs) executing internal and external clock synchronization using swarm consensus algorithms. NICs gather time signals from peers, aggregate pairwise error values, and adjust their clocks to achieve consensus, while external synchronization with UTC is achieved by probing time servers and propagating consensus times.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If PTP tree topology is used for clock synchronization, then clock synchronization can be achieved, but NIC-to-NIC synchronization accuracy deteriorates at scale due to node errors and jitter

Engineering Contradiction:
Improvefault toleranceVSAvoidNIC-to-NIC synchronization accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent divides the monolithic PTP synchronization system into independent NIC-level synchronization units. Each NIC independently synchronizes with its peer NICs using local swarm consensus, eliminating the single point of failure at the root time server and breaking the error propagation chain inherent in tree topology.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from the hierarchical tree topology (vertical dimension) to a mesh-like peer-to-peer topology (horizontal dimension). This dimensional shift allows multiple synchronization paths between any two NICs, providing both fault tolerance through redundancy and high accuracy through direct pairwise measurements without intermediate node errors.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Ease of operation

If PTP uses root time server and network switches for synchronization, then time distribution can be achieved, but synchronization accuracy deteriorates due to jittery root time server and hardware components

Engineering Contradiction:
Improvetime distribution capabilityVSAvoidsynchronization accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent extracts the time-keeping function from external infrastructure (root time server and network switches) and embeds it directly in each NIC. This extraction eliminates the jitter and errors introduced by external hardware components, as each NIC becomes its own time reference point while still participating in collective synchronization.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Each NIC performs its own synchronization measurements and computations independently without relying on the root time server or network switches. The NICs self-organize into swarms and perform local consensus algorithms, making the system autonomous and immune to external jitter and hardware failures.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If PTP tree topology is used, then clock synchronization can be established, but fault tolerance deteriorates due to brittle topology where node errors cause divergence

Engineering Contradiction:
Improveclock synchronizationVSAvoidfault tolerance
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent prepares multiple independent synchronization paths in advance through the mesh-like peer-to-peer topology. When a node fails or introduces errors, the system has pre-established alternative routes for time synchronization, cushioning against the impact of failures and preventing cascade divergence.

Inventive Principle:
Principle #11Beforehand cushioning (Prior cushioning)

Solution Approach 2:

The patent changes the topological parameter from hierarchical tree structure to distributed mesh structure. This parameter change fundamentally alters the system's fault tolerance characteristics, transforming it from brittle (single path failure causes divergence) to robust (multiple paths provide redundancy and error isolation).

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4554120A1Clock synchronization at scale in data centers
Publication Date: 2025.05.14 GOOGLE LLC
  • EP4554120A1 patent drawingFigure 1
  • EP4554120A1 patent drawingFigure 2
  • EP4554120A1 patent drawingFigure 3

AI summary

Aspects of the disclosed technology include techniques and mechanisms for performing clock synchronization at scale. A network device may gather, through repeated probe iterations to a swarm of peer network devices, a time indicated by each device. The network device may aggregate the gathered times to determine an offset and drift rate and may use one or more swarm consensus algorithms to determine a consensus time toward which the swarm may move. The swarm may synchronize to the consensus time. The network device may probe one or more time servers to retrieve a time signal indicated therein. The network device may propagate the retrieved time signal to the swarm. The swarm may move toward the time signal.