Replication Write Prioritization via Site-Level Queues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing replication configurations prioritize replicated writes based on per volume-level, which is insufficient and inadequate, as they do not account for the specific source site that sent the writes, leading to potential data loss and recovery point objectives (RPO) inconsistencies across different source sites.
Innovation Solution
Implementing a system that prioritizes replicated writes based on the source site level, combining site-level and storage resource level priorities, using a token-based approach to manage resource allocation and apportionment, allowing source sites to request elevated priorities dynamically in response to workload changes or RPO alerts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If per volume-level prioritization is used for replicated writes, then the replication configuration is simple to implement, but it cannot account for source site-specific requirements leading to data loss and RPO inconsistencies
Solution Approach 1:
The patent segments the prioritization mechanism into two distinct levels: site-level prioritization and volume-level prioritization. Site-level priorities are assigned to source sites to ensure critical sites receive adequate resource allocation, while volume-level priorities handle specific storage resource requirements. This segmentation allows the system to address both source site-specific requirements and individual volume needs, resolving the contradiction between reliability and complexity by organizing prioritization into manageable hierarchical layers.
Solution Approach 2:
The patent introduces a new dimension to the prioritization mechanism by adding site-level prioritization above the existing volume-level prioritization. This creates a two-dimensional prioritization framework where site-level priorities provide a macro-level view for resource allocation across multiple sources, while volume-level priorities provide micro-level control for specific storage resources. This dimensional expansion enables the system to account for source site requirements without completely redesigning the existing volume-level mechanism.
2Productivity
If site-level prioritization is implemented, then RPO management improves by prioritizing writes from critical source sites, but the system complexity increases with multiple priority queues and token management
Solution Approach 1:
The patent implements dynamic priority adjustment capabilities that allow source sites to request elevated priorities in response to workload changes or RPO alerts. The system can dynamically modify site-level priorities and reassign token allocations based on current conditions, enabling the prioritization mechanism to adapt to changing recovery requirements. This dynamic behavior improves recovery efficiency by allowing critical sites to receive increased resources when needed, while the system automatically adjusts to maintain balance across all sites.
Solution Approach 2:
The patent incorporates feedback mechanisms where source sites can monitor their RPO status and request priority adjustments when thresholds are breached. The system processes these requests and adjusts token allocations accordingly, creating a closed-loop control system. This feedback-driven approach enables the system to respond to actual performance conditions, improving recovery efficiency by automatically allocating more resources to sites experiencing RPO violations while maintaining overall system stability.
3Quantity of substance
If token-based resource allocation is used, then resource apportionment among multiple source sites is optimized, but the management and configuration of token distribution becomes more complex
Solution Approach 1:
The patent implements a token-based allocation system where the total token pool is apportioned among source sites according to their priority levels and resource requirements. Rather than attempting to perfectly balance all parameters, the system uses token allocations as a simplified metric that captures the essential resource distribution needs. This partial action approach optimizes resource apportionment by focusing on the most critical allocation decisions at the site level, while accepting that fine-grained configuration complexity is reduced in exchange for overall allocation efficiency.
Data Source
AI summary
Techniques can include: configuring source sites in a replication configuration with a target site; receiving, at a first of the source sites from a host, a write I/O to a target location of a source storage object of the source site; sending a replicated write corresponding to the write I/O from the first source site to the target site, wherein the replicated write writes to the target location of a target storage object of the target site; prioritizing the replicated write, based at least in part, on a site-level priority associated with the first source site; placing the first replicated write in a first queue of a first plurality of queues based, at least in part, on the site level priority associated with the first source site; selecting the first replicated write from said first queue; and servicing the first replicated write.