Distributed Storage Hash Function Selection via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional distributed storage systems suffer from reduced mapping speed and increased latency due to the use of static hash functions, leading to excessive resource consumption and frequent polling of storage devices.
Innovation Solution
A machine learning model dynamically selects a hash function from a set of candidate hash functions based on performance factors such as storage, network, and workload characteristics, allowing the distributed storage system to optimize its operation by adapting the hash function in response to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a static hash function is used to map data items to storage devices, then the system implementation is simple, but the mapping speed decreases and latency increases over time
Solution Approach 1:
The patent implements dynamic hash function selection by training a machine learning model to predict optimal hash functions based on system state features. The system transitions from using a fixed static hash function to dynamically selecting from multiple candidate hash functions, allowing the mapping speed to adapt to changing system conditions while maintaining implementation feasibility through automated model-based selection
Solution Approach 2:
The patent changes the parameter of hash function selection from static to dynamic by introducing a machine learning model that outputs different hash function choices based on learned patterns. The model analyzes system features and selects appropriate hash functions, effectively changing the hash function parameter adaptively rather than maintaining a constant parameter throughout system operation
2Device complexity
If a static hash function is used for data mapping, then the system configuration is simple, but processing time increases due to excessive polling
Solution Approach 1:
The patent applies preliminary action by pre-training a machine learning model offline to learn optimal hash function selection strategies. During runtime, the pre-trained model quickly predicts the best hash function based on current system state, avoiding the need for complex real-time analysis while reducing processing time and excessive polling behavior through informed pre-computed decisions
3Use of energy by moving object
If a static hash function is used, then resource consumption is high due to frequent polling, but implementing dynamic selection increases system complexity
Solution Approach 1:
The patent implements self-service by enabling the system to automatically select optimal hash functions through a trained machine learning model without requiring manual intervention or complex configuration management. The model autonomously analyzes system features and makes selection decisions, reducing computing resource consumption through intelligent automation while keeping operational complexity manageable through self-managing behavior
Data Source
AI summary
One example described herein includes a system that can receive values for performance factors associated with a distributed storage system. The system can provide the values as input to a machine-learning model configured to select a hash function from among a set of hash functions based on the values and generate an output indicating the hash function. The system can then configure the distributed storage system to map a data item to a storage device using the hash function.