Asynchronous Replication Mode Switching for Write Cache Overload
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing asynchronous replication techniques face inefficiencies and resource overheads when attempting to achieve very low Recovery Point Objectives (RPOs) such as near-zero or low RPOs, particularly due to excessive consumption of write tracking cache, leading to resource depletion and increased RPOs during high workloads.
Innovation Solution
A method is implemented to proactively demote volumes from a low RPO replication technique to a second asynchronous replication technique that does not utilize write tracking cache when resource consumption exceeds a high watermark threshold, based on priority, group membership, and cache consumption, ensuring efficient resource usage and maintaining RPO compliance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the first replication technique with write tracking cache optimization is used to achieve near-zero RPO, then replication performance and RPO compliance are improved, but write tracking cache consumption increases excessively leading to resource depletion
Solution Approach 1:
The system dynamically changes the replication mode parameter based on cache consumption levels. When write tracking cache consumption exceeds a threshold, the system transitions from the first replication technique (with write tracking cache) to the second replication technique (without write tracking cache), thereby adjusting the parameter to prevent resource depletion while maintaining operational continuity
Solution Approach 2:
The replication technique is made dynamic by enabling automatic demotion between first and second replication modes based on real-time cache consumption monitoring. This dynamic adjustment allows the system to adapt to varying workload conditions and cache availability, preventing resource exhaustion while optimizing replication performance
2Reliability
If multiple stretched volumes are configured for asynchronous replication with first replication technique, then replication coverage and data protection are improved, but total cache consumption increases leading to resource exhaustion during high workloads
Solution Approach 1:
Different replication techniques are applied to different stretched volumes based on their individual cache consumption characteristics and priority levels. High-priority volumes may continue using the first replication technique while lower-priority volumes are demoted to the second technique, allowing localized optimization without compromising overall data protection
Solution Approach 2:
The set of stretched volumes is segmented into different priority groups and replication modes. The system selectively applies the first replication technique to specific volumes based on their importance and cache consumption patterns, while applying the second technique to others, thereby distributing cache resource usage across different volume segments
Data Source
AI summary
Techniques can include: configuring stretched volumes to perform asynchronous replication in a first mode using a first replication technique that uses a write tracking cache to track locations that are written between successive replication-related snapshots of respective source volumes of the stretched volumes; performing asynchronous replication for the stretched volumes in accordance with the first replication technique; and monitoring resource consumption of the write tracking cache during asynchronous replication processing using the first replication technique, wherein said monitoring includes: determining that a current amount of write tracking cache consumed exceeds a high watermark threshold; and responsive to determining that the current amount of write tracking cache consumed exceeds a high watermark threshold, selecting, in accordance with criteria, one or more of the stretched volumes for demotion from the first replication mode to a second replication mode that uses a second replication technique that does not use the write tracking cache.