Decentralized Map-Reduce Communications via Direct Contacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-organizing linear communication orbits in distributed networks are not robust enough to handle errors efficiently, as error detection and recovery processes are confined to immediate neighboring nodes, leading to bottlenecks and inefficiencies in map-reduce communications.
Innovation Solution
Implementing a system where each node in the network can access other parts of the orbit through direct contacts, allowing for near-uniform distribution of load for organization, failure detection, and recovery, and using a responsibility hierarchy rule to dynamically assign nodes to orbits based on query similarity, enabling efficient and robust message communication and error handling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If nodes in a linear communication orbit only communicate with immediate neighboring nodes, then the system maintains simplicity in error detection and recovery, but the system becomes insufficiently robust and efficient when errors occur
Solution Approach 1:
The patent divides the linear communication orbit into segments by introducing direct contacts that can bypass intermediate nodes. When a node fails, the orbit is segmented into independent sections that can be recovered individually, improving robustness without requiring complete system restructuring.
Solution Approach 2:
The patent introduces direct contacts as intermediary nodes that provide alternative communication paths. These intermediaries enable error recovery by allowing nodes to communicate directly across failed segments, reducing dependency on immediate neighbors while maintaining overall system simplicity.
2Ease of operation
If a central management server is used to issue requests to all destination nodes, then centralized control is achieved, but the server becomes overwhelmed by communication load and creates a management bottleneck
Solution Approach 1:
The patent segments the centralized management function by distributing request issuance to multiple direct contacts throughout the orbit. This eliminates the single-point bottleneck while maintaining coordinated control, as each direct contact manages a specific segment of the network independently.
Solution Approach 2:
The patent transitions from a single-dimensional centralized control model to a multi-dimensional distributed control structure. Direct contacts operate at different positions in the orbit, creating a spatial distribution of management functions that improves throughput while maintaining control coherence.
3Reliability
If direct contacts are introduced to improve error recovery, then robustness increases, but the load distribution across nodes becomes uneven
Solution Approach 1:
The patent implements dynamic load distribution where direct contacts are selected and assigned responsibilities based on real-time network conditions and node capabilities. This dynamic assignment ensures even load distribution while maintaining the robust error recovery capabilities provided by direct contacts.
Solution Approach 2:
The patent changes the parameters of direct contact selection and responsibility assignment to optimize load distribution. By adjusting selection criteria and responsibility metrics, the system achieves both robust error recovery and uniform load distribution across all nodes.
Data Source
AI summary
A first machine identifies, from among a non-static collection of machines, a respective set of forward contacts that comprises a set of machines. The set of forward contacts are distributed along the ordered sequence in the forward direction away from the respective machine in an order of increasing similarity between the respective channel number assigned to the first machine and a respective channel number assigned to each of the set of forward contacts. The first machine establishes a respective direct communication channel between the first machine and each of the set of forward contacts. The first machine sends a first query to a first forward contact and sends collected answers for the first query to at least a second forward contact that has a greater similarity to the first machine based on the respective channel numbers of the first machine and the first and second forward contacts.