Programmable Network Interface Device for Vector Database High Availability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Maintaining high availability of sharded vector databases and virtual machines in AI infrastructure is challenging due to node unavailability, which disrupts the inference pipeline in RAG setups.

Innovation Solution

Utilizing programmable network interface devices, such as IPUs, DPUs, EPUs, and smart NICs, to manage replicas, provide a unified frontend, track heartbeats, load balance, mitigate node failures, and manage recovery and migration, ensuring dynamic and real-time replication of state across devices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If vector databases are distributed across multiple shards for load balancing, then productivity is improved, but reliability deteriorates due to node unavailability disrupting the inference pipeline

Engineering Contradiction:
Improveload balancingVSAvoidnode unavailability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements flow replication by creating copies of network flows between active and standby programmable network interface devices. When a node becomes unavailable, the standby device already has replicated flow state and can immediately take over, preventing disruption to the inference pipeline. This copying mechanism ensures reliability while maintaining the distributed sharded architecture for productivity.

Inventive Principle:
Principle #26Copying

2Reliability

If programmable network interface devices replicate state in real-time for high availability, then reliability is improved, but device complexity increases

Engineering Contradiction:
Improvehigh availabilityVSAvoidstate replication
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary replication mechanism where the active programmable network interface device automatically replicates flow state to a standby device through standardized protocols. This intermediary approach abstracts the complexity of real-time state synchronization, providing high availability without requiring complex custom replication logic at each device. The standby device simply receives and maintains replicated state, reducing overall system complexity.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If node failures are mitigated through replication, then reliability is improved, but loss of time increases due to replication and failover processes

Engineering Contradiction:
Improvefailure mitigationVSAvoidreplication and failover time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent implements preliminary action by pre-replicating flow state from active to standby programmable network interface devices before failures occur. The standby devices maintain ready-to-use replicated state in advance, so when a node fails, the failover is immediate without requiring time-consuming state synchronization during the failure event. This preliminary replication minimizes both the replication time overhead and failover time loss.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20250117673A1High availability ai via a programmable network interface device
Publication Date: 2025.04.10 INTEL CORP
  • US20250117673A1 patent drawing
  • US20250117673A1 patent drawing
  • US20250117673A1 patent drawing

AI summary

Techniques described herein address the above challenges that arise when using host executed software to manage vector databases by providing a vector database accelerator and shard management offload logic that is implemented within hardware and by software executed on device processors and programmable data planes of a programmable network interface device. In one embodiment, a programmable network interface device includes infrastructure management circuitry configured to facilitate data access for a neural network inference engine having a distributed data model via dynamic management of a node associated with the neural network inference engine, the node including a database shard of a vector database.