Multi-Node Storage I/O Redirection with Load-Aware Replies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing dual-node storage systems with active-active configurations face challenges in managing I/O operations, leading to increased resource contention, synchronization overhead, and performance bottlenecks due to internode communication and redirection of I/Os.
Innovation Solution
Implementing a method where the storage system internally maintains per-volume ownership, allowing hosts to access volumes from both nodes while redirecting I/Os from non-owner nodes to owner nodes, and adjusting the information returned based on interconnect load to minimize load on the interconnection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If I/O operations are redirected from non-owner nodes to owner nodes in active-active configuration, then resource contention and synchronization overhead are reduced, but interconnect load increases due to redirect requests and replies
Solution Approach 1:
The system dynamically adjusts redirect reply behavior based on real-time interconnect load conditions. When the interconnect is overloaded, the owner node sends only physical location information instead of actual data, allowing the non-owner node to fetch data directly from storage. This dynamic adaptation resolves the contradiction by making the redirection mechanism flexible and load-aware.
Solution Approach 2:
The system changes the parameter of redirect reply content based on interconnect load status. Under normal conditions, full data is transferred; under overload conditions, only metadata (physical location information) is transferred. This parameter change optimizes the balance between reducing resource contention and minimizing interconnect load.
2Device complexity
If per-volume ownership is maintained internally while allowing host access from both nodes, then synchronization overhead is reduced, but I/O redirection complexity increases
Solution Approach 1:
The storage system automatically manages I/O redirection based on internal ownership metadata without requiring explicit coordination between nodes. When a non-owner node receives an I/O request for a volume it doesn't own, it automatically redirects the request to the owner node using stored mapping information. This self-service mechanism reduces synchronization overhead while managing redirection complexity internally.
Solution Approach 2:
The system uses internal ownership metadata and mapping information as intermediaries to manage the redirection process. These intermediaries store the relationship between volumes and owner nodes, enabling automatic redirection decisions without complex synchronization protocols between nodes.
3Ease of operation
If redirect replies include full data content, then data access is simplified for the requesting node, but interconnect bandwidth consumption increases
Solution Approach 1:
The system applies partial action by transferring only the necessary information (physical location metadata) when interconnect load is high, rather than always transferring complete data. This selective data transfer approach reduces interconnect bandwidth consumption while maintaining ease of operation through automated redirection.
Solution Approach 2:
Instead of copying full data content across the interconnect, the system creates a lightweight copy of the physical location information in the redirect reply. The requesting node then uses this metadata to access the actual data directly from storage, reducing bandwidth requirements while maintaining functional equivalence.
Data Source
AI summary
Techniques can include: receiving, at a first node, a read I/O directed to a logical address; determining that the first node does not own the logical address; sending over an interconnect a redirect request from the first node to a second node that owns the first logical address; and the second node performing processing including: determining, in accordance with one or more criteria, whether the interconnect is overloaded; responsive to determining that the interconnect is not overloaded, sending a redirect reply over the interconnect to the first node where the redirect reply includes first content stored at the first logical address; and responsive to determining that the interconnect is overloaded, sending the redirect reply over the interconnect to the first node where the redirect reply includes physical location information identifying a first physical storage location or address on non-volatile storage of the first content of the first logical address.