Distributed Data Replica Placement via Hierarchical Tree Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed data storage systems without a central node face reliability issues and inefficient data migration when the number of nodes changes, leading to reduced system performance and increased data loss.
Innovation Solution
A method for storing data replicas in a distributed data storage system with a tree structure, where child nodes are selected level by level from intermediate nodes using a preset mapping mode, with the mapping mode reset for each selection, ensuring target storage positions are dispersed and different for each replica, thereby improving reliability and reducing data migration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data replicas are stored using traditional hash function with modulo operation in a distributed data storage system without center nodes, then the system avoids central node bottleneck, but data migration occurs extensively when the number of nodes changes
Solution Approach 1:
The patent introduces a hierarchical tree structure with multiple levels (machine room level, rack level, host level, storage device level) to replace the traditional flat hash-modulo addressing scheme. This dimensional transformation allows data location to be determined by tree traversal rather than direct modulo calculation, enabling selective migration only of affected data paths when nodes are added or removed.
Solution Approach 2:
The distributed storage system is segmented into a hierarchical tree structure where data is distributed across multiple levels of nodes. Each level (machine room, rack, host, storage device) represents a segmentation layer, allowing independent management and minimal cross-level migration when changes occur at any single level.
2Device complexity
If all data replicas are stored in hard disks of hosts within the same rack, then the system simplifies storage management, but reliability is reduced when power or network failure occurs in that rack
Solution Approach 1:
The patent implements local quality by distributing data replicas across different hierarchical levels and spatial locations. Instead of uniform storage in a single rack, data is placed in specific hosts and storage devices selected through tree traversal, ensuring that replicas are geographically and hierarchically dispersed to avoid single-point failures.
Solution Approach 2:
The system preemptively distributes data replicas across multiple racks and machine rooms through the hierarchical tree structure before failures occur. This prior cushioning ensures that if one rack experiences power or network failure, other replicas in different physical locations remain accessible, maintaining system availability.
3Ease of manufacture
If data replicas are distributed randomly on storage devices using hash function, then the system achieves simple data placement, but data migration is inevitable when the number of nodes changes
Solution Approach 1:
The patent transforms the flat random distribution approach into a hierarchical tree-based placement system. Data location is determined by traversing the tree structure from machine room level down to storage device level, providing structured distribution that maintains simplicity while enabling selective migration based on tree path changes rather than complete redistribution.
4Reliability
If a center node is used to manage data replica positions, then data placement is optimized according to load conditions, but the center node becomes a bottleneck and single point of failure
Solution Approach 1:
The patent implements self-service by enabling each node in the distributed system to independently determine data replica positions through local tree structure information and hash calculations. Nodes autonomously traverse the hierarchical tree and calculate storage locations without querying a central coordinator, eliminating the performance bottleneck while maintaining optimized placement based on distributed load information.
Data Source
AI summary
Provided are a data replica storage method, a terminal apparatus, and a server apparatus, which help to improve the reliability of a distributed data storage system without a central node and reduce an amount of migrated data when the number of nodes of the system changes. The method comprises: for each data replica to be stored, selecting child nodes level by level from a pre-selected intermediate node until a leaf node according to a tree-structured a distributed data storage system without a central node, and determining the leaf node as a target storage position of the data replica, selection of a child node comprising: combining an identification of data to be stored and an identification of a child node into a host image, obtaining a mapping value related to the child node using a preset mapping mode, selecting a sub-node of a central node according to the mapping value, and resetting a mapping relationship before each time when the child node is selected; and storing all data replicas to be stored at the target storage positions of the data replicas.