Accelerator Checkpointing With Bit-Vector Parallel Recovery

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing checkpointing methods for fault tolerance in applications using accelerators often cause performance degradation due to delayed checkpointing and task execution delays, as they cannot perform tasks in parallel during the checkpointing process.

Innovation Solution

A method and system that utilize a bit vector to determine when checkpointing is completed for each accelerator data, allowing operations to proceed while checkpointing is performed, by prioritizing operations on completed checkpointed data and storing accelerator data in the order of write access.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If checkpointing is performed by storing accelerator data at a specific time point, then fault tolerance is improved, but task execution is delayed because other tasks cannot be performed in parallel during checkpointing

Engineering Contradiction:

Improvefault toleranceVSAvoidtask execution speed

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent segments the checkpointing process by introducing a bit vector to track the checkpointing status of each accelerator data independently. This allows the system to identify and operate on accelerator data that has completed checkpointing while others are still being checkpointed, enabling parallel task execution alongside checkpointing operations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary actions by maintaining a bit vector that records the checkpointing completion status of each accelerator data before task execution. This preliminary tracking enables the system to determine which accelerator data is ready for operation, allowing tasks to proceed without waiting for all checkpointing to complete, thus resolving the contradiction between fault tolerance and task execution speed.

Inventive Principle:

Principle #10Preliminary action

2Reliability

If checkpointing is performed after accelerator operation completes, then fault tolerance is maintained, but checkpointing is delayed causing performance degradation

Engineering Contradiction:

Improvefault toleranceVSAvoidcheckpointing delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent introduces dynamic status tracking through a bit vector that reflects the real-time checkpointing completion status of each accelerator data. This dynamic information allows the system to flexibly determine when operations can proceed, transforming the static sequential checkpointing process into a dynamic parallel process that reduces checkpointing delays while maintaining fault tolerance.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The bit vector serves as a feedback mechanism that continuously provides information about checkpointing completion status to the task scheduler. This feedback enables the system to make informed decisions about task execution timing, allowing operations to start as soon as their required accelerator data is checkpointed, thereby reducing overall checkpointing delay while preserving fault tolerance capabilities.

Inventive Principle:

Principle #23Feedback

3Ease of operation

If task execution command is issued during checkpointing, then system responsiveness is improved, but task execution is delayed until checkpointing completes

Engineering Contradiction:

Improvesystem responsivenessVSAvoidtask execution delay

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The patent segments the system response to task execution commands by using the bit vector to identify which accelerator data is ready for operation. When a task execution command is received, the system can immediately proceed with operations on accelerator data that has completed checkpointing while other data continues to be checkpointed in the background, thus improving system responsiveness without sacrificing fault tolerance.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The bit vector performs preliminary tracking of checkpointing completion status, enabling the system to respond immediately to task execution commands for ready data. This preliminary information availability eliminates the need to wait for checkpointing completion before executing tasks, reducing execution delay while maintaining the ability to restore from checkpoints if failures occur.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12561121B2Methods and system for performing checkpointing for fault tolerance in application

Publication Date: 2026.02.24 MOREH CORP

AI summary

A method for performing checkpointing for fault tolerance of an application is provided, which is performed by one or more processors, and includes storing accelerator data associated with an application at a specific time point as a checkpoint, storing operation data associated with the application performed after the specific time point, and performing application fault tolerance based on the checkpoint and the stored operation data.