Storage Cache Manager Tagging for IO Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The movement of snapshots by the cloud tethering subsystem in storage systems can skew the algorithms used by the cache manager and compression subsystem, leading to reduced performance and inaccurate prediction of future IO operations, as it involves long sequential read operations and consumes space in the uncompressed data pool.
Innovation Solution
By tagging thin devices used for snapshot movement to the cloud repository, the cache manager and compression subsystem are made aware to exclude data read operations from these devices from their algorithms and data pools, preventing skewing and conserving space, thus maintaining accurate prediction of future IO operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the cache manager includes all data read operations in its prediction algorithms, then the algorithms can capture all IO patterns, but snapshot movement operations skew the prediction accuracy
Solution Approach 1:
The patent segments data read operations into two categories: those from tagged thin devices (snapshot movements) and those from untagged thin devices (normal operations). The cache manager applies different handling rules to each segment, excluding tagged operations from prediction algorithms while including untagged operations, thereby maintaining prediction accuracy without being skewed by snapshot movements.
Solution Approach 2:
The patent introduces a tagging mechanism as an intermediary identifier on thin devices. This tag acts as a mediator that allows the cache manager to distinguish between snapshot movement operations and normal IO operations. The tagging system enables selective exclusion of snapshot operations from prediction algorithms without affecting the processing of normal operations.
2Speed
If the compression subsystem includes all data in the uncompressed data pool, then all data can be accessed quickly, but snapshot movement consumes excessive space
Solution Approach 1:
The patent segments the uncompressed data pool into two segments: data from tagged thin devices (snapshot movements) and data from untagged thin devices (normal operations). The compression subsystem excludes data from tagged devices from the uncompressed data pool, maintaining space efficiency while preserving quick access for normal operations through selective compression.
Solution Approach 2:
The patent extracts data from tagged thin devices (snapshot movements) out of the normal uncompressed data pool management flow. By identifying and separating these operations through tagging, the system prevents snapshot movement data from consuming valuable uncompressed pool space, thereby conserving resources for actual operational data.
3Device complexity
If the system processes all read operations uniformly, then the processing logic remains simple, but snapshot movements skew cache management algorithms
Solution Approach 1:
The patent applies preliminary action by tagging thin devices before snapshot movement operations occur. This pre-tagging allows the cache manager to identify and exclude snapshot operations from prediction algorithms without requiring complex real-time analysis. The preliminary tagging simplifies the processing logic while maintaining accurate cache management.
Data Source
AI summary
A method of increasing the accuracy of predicting future IO operations on a storage system includes creating a snapshot of a production volume, linking the snapshot to a thin device, mounting the thin device in a cloud tethering subsystem, and tagging the thin device to identify the thin device as being used by the cloud tethering subsystem. When data read operations are issued by the cloud tethering subsystem on the tagged thin device, the data read operations are executed by a front-end adapter of the storage system to forward data associated with the data read operations to a cloud repository. The cache manager, however, does not use information about data read operations on tagged thin devices in connection with predicting future IO operations on the cache, so that movement of snapshots to the cloud repository do not skew the algorithms being used by the cache manager to perform cache management.