Memory Snapshot Checkpointing for Low-Idle Application State Copy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory architectures struggle to efficiently manage checkpoint copies of application states due to resource-intensive snapshotting processes that cause significant idle time and resource contention, particularly in high-performance computing environments where low job stop-time tolerances are required.
Innovation Solution
Utilizing shared memory snapshots on fabric-attached memory devices within a CXL architecture to create snapshots of data objects, allowing concurrent processing and modification while isolating producer and consumer access, thereby reducing idle time and enabling efficient checkpoint copying.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional snapshotting processes are used to create checkpoint copies, then data consistency is ensured, but application idle time increases significantly and resource contention occurs
Solution Approach 1:
The system performs preliminary actions by continuously maintaining write-enabled snapshots in the background before checkpointing is actually needed. This allows the snapshot to be ready in advance, eliminating the need to pause the application when a checkpoint is requested. The snapshot is updated incrementally as data changes, so when checkpointing occurs, the copy is already prepared and can be immediately used.
Solution Approach 2:
The system enables continuous useful action by allowing the application to keep writing to and modifying data during the snapshot process. Unlike traditional methods that require stopping writes, this approach maintains continuous data modification while simultaneously maintaining a consistent copy for checkpointing purposes, thus eliminating idle time while preserving data consistency.
2Measurement precision
If traditional snapshotting processes are used to create checkpoint copies, then accurate program state capture is achieved, but system resource contention increases
Solution Approach 1:
The system uses copying by creating a duplicate copy of the data for the snapshot that can be independently maintained. This copy is kept synchronized with the original data through incremental updates, allowing the original data to be freely modified without affecting the snapshot's accuracy. The copy mechanism enables accurate program state capture while reducing resource contention since modifications to the original don't block snapshot operations.
3Productivity
If rapid snapshot creation is implemented to reduce idle time, then application throughput improves, but snapshot consistency may be compromised
Solution Approach 1:
The system performs preliminary action by pre-establishing the snapshot structure and continuously updating it in the background before rapid checkpointing is needed. This preliminary preparation allows the system to quickly create consistent snapshots when required, because the foundational work of tracking and synchronizing data changes has already been completed incrementally, not rushed at the moment of checkpointing.
Solution Approach 2:
The system uses an intermediary mechanism (the continuously maintained snapshot copy) that mediates between the need for rapid checkpointing and snapshot consistency. This intermediary copy acts as a buffer that can be quickly copied for checkpointing while its consistency is maintained through background synchronization processes, thus enabling both speed and reliability.
Data Source
AI summary
A system can include a memory device and a processing device operatively coupled with the memory device, to perform operations of maintaining a set of data items reflective of a state of a host application and receiving a first request to make a checkpoint copy of the state of the host application. The operations can also include responsive to receiving the first request, associating a snapshot identifier with the object identifier, and receiving a second request to modify a data item of the set of data items at the first set of physical address ranges. They can also include executing the second request by recording a modification a data item exclusively at a second set of physical address ranges and making the checkpoint copy by copying the set of data items of the snapshot from the first set of physical address ranges to a second memory device.