Predictive Garbage Collection for SMR Storage

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current storage systems, particularly those using shingled magnetic recording (SMR) disks, face high computational and operational costs during garbage collection due to the need to rewrite entire zones when altering individual files, and existing methods do not efficiently determine which files to delete, leading to excessive data processing and resource utilization in data centers.

Innovation Solution

The method involves predicting the survivor lifetime of individual files within a container using machine learning algorithms to determine a deletion predictor, calculating the cost and benefit of garbage collection, and scheduling it at optimal epochs where the net benefit is positive, thereby optimizing garbage collection operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If garbage collection is performed on SMR disks by rewriting entire zones, then storage density is increased, but computational cost and operational expense increase significantly

Engineering Contradiction:
Improvestorage densityVSAvoidcomputational cost
Core Design Contradiction:
Quantity of substanceVSLoss of energy

Solution Approach 1:

The patent segments the container into individual files and predicts which specific files are likely to be deleted using machine learning. Instead of treating the entire zone as a single unit that must be rewritten, the system identifies and processes only the relevant files for garbage collection, thereby reducing the computational cost while maintaining the storage density benefits of SMR disks.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If the entire collection of data is read and processed to identify files for deletion, then accurate garbage identification is achieved, but operational cost and processing time increase

Engineering Contradiction:
Improvegarbage identification accuracyVSAvoidoperational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies preliminary action by using machine learning models to predict which files are likely to be deleted before garbage collection is performed. This prediction step is executed in advance, allowing the system to identify target files without reading and processing the entire collection of data, thus improving operational efficiency while maintaining accurate garbage identification.

Inventive Principle:
Principle #10Preliminary action

3Loss of energy

If frequent garbage collection is performed to maintain storage efficiency, then data storage costs are reduced, but resource utilization and processing overhead increase

Engineering Contradiction:
Improvedata storage costVSAvoidresource utilization
Core Design Contradiction:
Loss of energyVSUse of energy by moving object

Solution Approach 1:

The patent implements dynamic garbage collection scheduling based on predicted file deletion patterns. The machine learning model continuously learns from access patterns and deletion behaviors, adjusting the timing and scope of garbage collection operations dynamically. This allows the system to perform garbage collection at optimal moments when it provides maximum benefit with minimum resource utilization, rather than following a fixed frequent schedule.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3923130B1Optimizing garbage collection based on survivor lifetime prediction
Publication Date: 2024.04.17 GOOGLE LLC
  • EP3923130B1 patent drawingFigure 1
  • EP3923130B1 patent drawingFigure 2
  • EP3923130B1 patent drawingFigure 3A

AI summary

A predictive method for scheduling of the operations is described. The predictive method utilizes data generated from computing an expected lifetime of the individual files or objects within the container. The expected lifetime of individual files or objects can be generated based on machine learning techniques. Operations such as garbage collection are scheduled at an epoch where computational efficiencies are realized for performing the operation.