Neural Network I/O Scheduling for Storage Background Tasks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage systems face performance issues due to background tasks like garbage collection and wear levelling, which cause increased latency and are protocol-dependent, affecting overall system efficiency.
Innovation Solution
A machine-learning-based approach using a neural network circuit to predict background tasks, enabling smart re-routing and I/O buffering to minimize latency and improve throughput without modifying existing interfaces or hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If background tasks like garbage collection and wear levelling are performed in storage devices, then storage device maintenance and reliability are improved, but system latency increases and overall system efficiency deteriorates
Solution Approach 1:
The neural network predicts future background tasks before they occur, allowing the I/O scheduler to proactively reroute I/O requests to alternative storage devices or buffer them in memory. This preliminary action prevents latency spikes by avoiding conflicts between background tasks and I/O operations, resolving the contradiction between maintaining storage reliability and minimizing system latency
Solution Approach 2:
The patent introduces an I/O scheduler as an intermediary layer between the host and storage devices. This scheduler receives I/O requests, consults neural network predictions about background tasks, and intelligently reroutes requests to alternative storage devices or buffers them in memory. The intermediary enables coordination between background maintenance tasks and I/O operations, allowing both storage reliability and low latency to be achieved simultaneously
2Device complexity
If traditional I/O scheduling methods are used, then system simplicity is maintained, but protocol dependency increases and adaptability deteriorates
Solution Approach 1:
The I/O scheduler acts as a protocol-agnostic intermediary that sits between the host and storage devices. It uses neural network predictions to make intelligent routing decisions without being tied to any specific storage protocol. This intermediary layer provides adaptability to different protocols while maintaining relative system simplicity by using a unified scheduling approach across multiple storage devices
Solution Approach 2:
The patent implements a universal I/O scheduling mechanism that can handle multiple storage protocols and device types through a single neural network-based prediction system. The scheduler's ability to reroute I/O requests to alternative storage devices makes the system protocol-independent and adaptable to different storage configurations, achieving versatility without proportionally increasing complexity
Data Source
AI summary
A method for scheduling requests in a storage system includes receiving, at a scheduler, a request associated with a first storage device, receiving, at the scheduler, task information from a neural network circuit, the task information being associated with a task performed by the first storage device, and modifying a processing of the request at the first storage device based on the task information.