Adaptive Data Replication Mode Switching for Lower Storage Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional storage systems face inefficiencies in data management and reliability due to unnecessary write operations and lack of centralized control over storage processes, leading to increased latency and reduced reliability.
Innovation Solution
Implementing a direct-mapped flash storage system where the operating system directly manages data blocks without address translation by storage controllers, offloading device management responsibilities, and utilizing non-volatile RAM for quick data buffering and proactive data rebuilding across storage nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If storage controllers perform address translation and device management, then data access flexibility is improved, but system latency increases and reliability decreases
Solution Approach 1:
The patent extracts the address translation and device management functions from the storage controllers and relocates them to the operating system. This removes the bottleneck at the controller level, allowing direct OS-managed access to flash storage devices without intermediary translation overhead, thereby reducing latency while maintaining flexibility through OS-level management.
Solution Approach 2:
The patent introduces a new intermediary layer in the form of OS-managed translation layers that handle address translation directly. This mediator operates at the OS level rather than at the controller level, enabling more efficient translation processes and reducing the time penalty associated with traditional controller-based translation.
2Device complexity
If storage controllers manage data blocks, then device control is simplified, but system reliability decreases due to single points of failure
Solution Approach 1:
The patent segments the data management responsibilities by dividing them between the operating system and multiple storage nodes. Instead of concentrating all management functions in single controllers, the OS directly manages data blocks across distributed storage nodes, eliminating single points of failure and improving system reliability through distributed architecture.
Solution Approach 2:
The patent implements self-service mechanisms where storage nodes perform proactive data rebuilding and recovery operations autonomously. When data corruption or node failure is detected, the system automatically initiates recovery processes without requiring controller intervention, thereby maintaining reliability while reducing management complexity.
3Quantity of substance
If traditional write operations are performed, then data is stored, but redundant writes increase latency and reduce efficiency
Solution Approach 1:
The patent implements preliminary actions by performing data validation, checksum verification, and recovery preparation before actual write operations complete. The system proactively prepares recovery data and validates integrity beforehand, preventing the need for redundant corrective writes later, thereby improving write operation efficiency without compromising data storage capacity.
4Reliability
If reactive data recovery is used, then data can be restored after failure, but latency increases and reliability is reduced
Solution Approach 1:
The patent applies preliminary action by implementing proactive data rebuilding that continuously maintains redundant copies and validates data integrity before failures occur. This advance preparation eliminates the need for time-consuming reactive recovery operations, reducing recovery latency while maintaining strong data recovery capabilities.
Solution Approach 2:
The patent implements beforehand cushioning through proactive data rebuilding mechanisms that create buffer copies and validation checksums in advance. This cushioning layer of pre-prepared recovery data protects the system against failures, enabling rapid restoration without the latency penalty of post-failure reconstruction.
Data Source
AI summary
A uniform model for distinct types of data replication, including receiving, at a source data repository, an update to a dataset; generating, based on the update to the dataset, both metadata describing the update to the dataset and also a metadata representation of the dataset; and initiating, based on the same metadata describing the update to the dataset and also based on the same metadata representation of the dataset, either a first type of data replication or a second type of data replication from among a plurality of types of data replication.