Clustered Filesystem Heatmap Adaptive Replication

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current distributed filesystems face issues such as slow access for non-local files, inability to track access patterns, limited POSIX compliance, inefficient replication, and lack of adaptability to access patterns, leading to suboptimal performance and reliability in managing data across multiple storage nodes.

Innovation Solution

A clustered filesystem that uses a heatmap of access activity to dynamically adjust the number of replicate nodes for data blocks, ensuring data locality and redundancy while maintaining filesystem consistency, and employing techniques like RDMA for efficient data transfer and replication.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If files are replicated to multiple nodes in a distributed filesystem, then data availability and storage capacity are improved, but access latency for non-local files increases and system complexity increases

Engineering Contradiction:
Improvedata availabilityVSAvoidaccess latency
Core Design Contradiction:
ReliabilityVSSpeed

Solution Approach 1:

The system proactively replicates data blocks to multiple nodes before they are actually needed, based on predicted access patterns. This preliminary replication ensures that when a node needs data, local copies are already available, reducing access latency while maintaining high data availability across the distributed filesystem

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates local copies of data blocks on specific nodes based on their access patterns and proximity to data sources. By making data locally available at the quality level needed (local copies rather than full replication everywhere), the system improves access speed for non-local files while maintaining reliability through selective replication

Inventive Principle:
Principle #3Local quality

2Quantity of substance

If data is replicated to multiple nodes, then storage capacity and availability are improved, but the ability to track access patterns and maintain coherence deteriorates

Engineering Contradiction:
Improvestorage capacityVSAvoidaccess pattern tracking
Core Design Contradiction:
Quantity of substanceVSLoss of information

Solution Approach 1:

The system implements feedback mechanisms where nodes report their access patterns, data block usage, and replication status back to the central coordinator. This feedback loop enables the system to track access patterns across all replicated nodes, maintain coherence information, and dynamically adjust replication strategies based on observed behavior while preserving storage capacity benefits

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system creates a universal view of the distributed filesystem where all nodes can access and track data blocks regardless of their physical location. The replicated data structures provide multi-functionality by serving both as storage containers and as tracking mechanisms for access patterns, eliminating the trade-off between storage capacity and information tracking

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Ease of manufacture

If whole files are replicated regardless of changed data blocks, then replication simplicity is improved, but replication efficiency and network bandwidth usage deteriorate

Engineering Contradiction:
Improvereplication simplicityVSAvoidreplication efficiency
Core Design Contradiction:
Ease of manufactureVSProductivity

Solution Approach 1:

The system segments files into smaller data blocks that can be independently tracked and replicated. Instead of replicating entire files, only the specific data blocks that have changed are identified and replicated to target nodes. This segmentation maintains replication simplicity through modular operations while dramatically improving replication efficiency by transmitting only necessary data portions

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system extracts only the changed portions of data blocks from source files and replicates these extracted segments to destination nodes. By taking out and replicating only the necessary changed data rather than whole files, the system maintains the simplicity of automated replication while achieving high replication efficiency and optimal network bandwidth utilization

Inventive Principle:
Principle #2Taking out (Extraction)

4Adaptability or versatility

If distributed filesystems use loose cluster coherence, then cluster expandability is improved, but the ability to prevent simultaneous modifications and maintain data consistency deteriorates

Engineering Contradiction:
Improvecluster expandabilityVSAvoiddata consistency
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system introduces an intermediary layer of replicated data structures and coordination protocols that mediate between the loosely connected cluster nodes. This intermediary mechanism enables nodes to expand the cluster freely while the intermediary layer maintains data consistency by coordinating modifications, tracking access patterns, and preventing conflicting simultaneous changes without requiring tight coupling between all nodes

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4176345B1Fully coherent efficient non-local storage cluster file system
Publication Date: 2024.12.11 ORACLE INT CORP
  • EP4176345B1 patent drawingFigure 1
  • EP4176345B1 patent drawingFigure 2
  • EP4176345B1 patent drawingFigure 3

AI summary

A heatmap is used to identify access patterns of ranges of data blocks at specific times, in order to optimize the size of the set of replicate nodes, for the purpose of reducing access latency. In an embodiment, the heatmap is used to enforce minimum replication of each data block such as when a replicate node crashes and is replaced. In an embodiment, the heatmap is used to adjust the minimum replication of a data block such as during and after a demand spike. In an embodiment, each data block is replicated on a minimum amount of respective replicate nodes of a cluster. A first node requests access to data blocks. Based on the requesting the access, a heatmap is modified, and the data blocks are replicated to the first node. Based on the heatmap, the minimum amount of nodes in the respective replicate nodes for at least one data block is adjusted.