Host Cache Coherency via Purge Counter Synchronization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Maintaining host cache coherency in data storage networks introduces additional delays and computational costs, negating the benefits of faster read response times provided by host caching, especially when multiple host systems share data.
Innovation Solution
Implementing a mechanism where each host system directly connects to the internal fabric of the storage system, allowing for direct communication with the storage system to determine data versioning and purge cache entries, and using a purge counter to synchronize cache updates across host systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If host cache coherency mechanisms are implemented to maintain data consistency across multiple host systems, then data consistency is improved, but response time delays and computational costs increase
Solution Approach 1:
The storage system proactively purges cache entries at the host system before data modification instructions are executed. By incrementing a purge counter and having host systems periodically check this counter, the storage system ensures cache invalidation occurs in advance, eliminating the need for hosts to wait for coherency checks during data modifications.
Solution Approach 2:
The storage system implements a feedback mechanism where it monitors the purge counter value and sends notifications to host systems when the counter changes. Host systems periodically poll the purge counter and compare it against their cached values, creating a feedback loop that maintains cache coherency without requiring continuous active monitoring or blocking operations.
2Reliability
If host cache coherency mechanisms are implemented to maintain data consistency across multiple host systems, then data consistency is improved, but computational costs increase
Solution Approach 1:
Host systems autonomously manage their own cache coherency by periodically checking the purge counter value received from the storage system. Each host independently compares the current purge counter against its cached value and purges entries as needed, eliminating the need for complex centralized coherency management protocols and reducing overall computational overhead.
Solution Approach 2:
Instead of continuously monitoring for data modifications, host systems periodically check the purge counter at intervals. This periodic polling approach significantly reduces computational costs compared to continuous monitoring while still ensuring cache coherency is maintained, as the intervals can be optimized based on workload characteristics.
3Reliability
If traditional cache coherency protocols are used where hosts communicate through storage system components, then data consistency is maintained, but response time delays increase due to intervening components
Solution Approach 1:
The host system's cache coherency management is segmented into independent components: the storage system generates and broadcasts purge counter updates, while each host system independently processes these updates and manages its own cache. This segmentation eliminates the need for hosts to communicate through intervening storage system components, allowing direct and faster cache invalidation operations.
Solution Approach 2:
The cache coherency management function is extracted from the storage system's internal fabric and moved to the host systems themselves. By having hosts directly receive purge counter updates from the storage system and autonomously manage their caches, the patent removes intervening components that would otherwise be required to mediate communication between hosts and storage, significantly improving response time.
Data Source
AI summary
A storage system may maintain a purge counter for one or more logical storage units. When an instruction is received to perform an operation that will modify data across the one or more logical storage units, the purge counter may be incremented. One or more host systems implementing host caching may periodically poll the storage system to determine the purge counter value. When the current value of the purge counter value is different than a previously polled purge counter value recorded on a host system, the host system may purge from its host cache any entries for logical storage units associated with the purge counter. The data storage system may not execute the data modification instruction until it receives acknowledgement from all host systems caching data affected by the modification instruction that the host system has purged any host cache entries corresponding to the LSUs affected by the modification operation.