Asynchronous Replication Write Concurrency Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for controlling write order in data storage systems that implement asynchronous replication fail to maintain efficient write ordering at remote sites, leading to latency issues and compromised data integrity, as they typically require serial issuance of writes, which limits performance and causes replicas to lag behind primary data.
Innovation Solution
The method involves using concurrency group identifiers to group and manage writes at the primary site, allowing parallel issuance of replica writes at the remote site, ensuring that write order is maintained and latency is reduced by reflecting the concurrency of writes as they occur on the primary site.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If serial issuance of writes is used at the remote site, then write order is maintained, but backup performance deteriorates and latency increases
Solution Approach 1:
The patent segments writes into concurrency groups based on their dependency relationships. Writes that can be executed in parallel are grouped together and assigned the same concurrency group identifier, while writes that must be sequential receive different identifiers. This segmentation allows the remote site to process multiple writes concurrently within each group while maintaining order between groups, resolving the contradiction between write order maintenance and backup performance.
2Speed
If asynchronous replication is used, then network latency is reduced and application speed improves, but write order control at the remote site deteriorates
Solution Approach 1:
The patent introduces concurrency group identifiers as an intermediary mechanism between the primary and remote sites. These identifiers are embedded with each write operation and serve as a mediator that carries ordering information across the asynchronous replication link. The remote site uses these identifiers to reconstruct the write order without requiring synchronous acknowledgment, thus maintaining both high speed and reliable write order control.
3Productivity
If parallel issuance of writes is used at the remote site, then backup performance improves, but write order integrity deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the remote site monitors the completion status of writes within each concurrency group. The system only allows progression to the next concurrency group after all writes in the current group have been successfully processed. This feedback loop ensures that parallel issuance of writes does not compromise write order integrity, as the system dynamically adjusts based on completion status while maintaining high backup performance.
Data Source
AI summary
A method for controlling write order in a remote data storage system used for asynchronous replication. The data backup method includes receiving writes issued concurrently by a host prior to any being completed. An async write manager, such as a software application on a storage controller or on a host, operates to group each of the writes into a concurrency group with a replica write corresponding to each of the host writes being queued for transmittal over a link to a remote data storage system. The replica writes each include a concurrency group identifier, such as a group number or last completed write number, that identifies which concurrency group the replica has been assigned. The method includes transferring the replica writes to the remote data storage system and processing the concurrency group identifiers with a write order manager to determine which subsets of the replica writes to issue concurrently.