GPU File Server Architecture for Distributed Metadata Scalability

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional file systems face scalability limitations due to the use of centralized metadata management, leading to potential data loss in case of server failure and inability to scale beyond small clusters.

Innovation Solution

A distributed file system architecture utilizing GPUs and non-volatile memory (NVM) that distributes metadata across multiple servers, enabling high-performance parallel commits and load balancing through penta-groups and distributed erasure coding, allowing for scalability to thousands of nodes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If centralized metadata management is used in conventional file systems, then simplicity of architecture is maintained, but scalability is limited and data loss risk increases

Engineering Contradiction:
Improvearchitecture simplicityVSAvoidscalability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent segments the centralized metadata management into distributed metadata servers organized in penta-groups. Each metadata server handles a portion of the namespace, allowing the system to scale to thousands of nodes while maintaining manageable complexity through modular organization and automated failure detection mechanisms.

Inventive Principle:
Principle #1Segmentation

2Ease of operation

If centralized metadata management is used, then ease of operation is maintained, but reliability deteriorates due to potential data loss on server failure

Engineering Contradiction:
Improveoperational simplicityVSAvoiddata safety
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The patent implements beforehand cushioning through automated failure detection and data redistribution mechanisms. When a metadata server fails, the system automatically detects the failure, redistributes the failed server's namespace portions to surviving members of the penta-group, and maintains data availability without requiring manual intervention, thus preserving ease of operation while enhancing reliability.

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

3Reliability

If metadata is distributed across multiple servers, then scalability and reliability are improved, but system complexity increases

Engineering Contradiction:
Improvefailure resilienceVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the functions of multiple metadata servers into organized penta-groups where five servers work together as a unified failure-resilient unit. This merging approach allows the system to achieve failure resilience through distribution while managing complexity through standardized group configurations and automated coordination protocols that handle failover and data redistribution transparently.

Inventive Principle:
Principle #5Merging (Combining)

4Reliability

If distributed architecture with penta-groups is implemented, then failure resilience is improved, but device complexity increases

Engineering Contradiction:
Improvefailure resilienceVSAvoidarchitecture complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements dynamics through automated failure detection and adaptive redistribution mechanisms. The penta-groups dynamically adjust their composition and data distribution based on server availability, with automated protocols that detect failures and redistribute namespace portions in real-time. This dynamic behavior provides failure resilience while managing complexity through automation rather than static, manually-configured systems.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3811324B1GPU based server in a distributed file system
Publication Date: 2025.11.26 WEKA IO LTD
  • EP3811324B1 patent drawingFigure 1
  • EP3811324B1 patent drawingFigure 2
  • EP3811324B1 patent drawingFigure 3

AI summary

A plurality of computing devices are communicatively coupled to each other via a network, and each of the plurality of computing devices is operably coupled to one or more of a plurality of storage devices. A plurality of failure resilient stripes is distributed across the plurality of storage devices such that each of the plurality of failure resilient stripes spans a plurality of the storage devices. A graphics processing unit is operable to access data files from the failure resilient stripes, while bypassing a kernel page cache. Furthermore, these data files may be accessed in parallel by the graphics processing unit.