Error Communication Logical Port for InfiniBand Storage Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
InfiniBand storage networks face challenges in detecting errors in communication partner Queue Pairs (QPs) without relying on timeout values, leading to delayed failure detection and potential host timeouts, which is critical for maintaining high reliability and prompt data transfer in storage devices.
Innovation Solution
The implementation of an error communication logical port in storage nodes to notify other nodes of a data communication port's error state, allowing for immediate detection and notification of QP errors, enabling faster failure handling and alternative path utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If timeout-based error detection is used in InfiniBand storage networks, then error detection can be implemented with existing HCA mechanisms, but error detection time is delayed (0.14 to 0.5 second) causing host timeouts
Solution Approach 1:
The patent implements preliminary action by having the error communication logical port proactively notify other storage nodes of QP error states before timeout occurs. When a QP transitions to error state, the error communication logical port immediately sends notification messages to partner nodes, eliminating the need to wait for timeout-based detection and enabling immediate failure response.
2Ease of manufacture
If traditional HCA timeout mechanisms are used, then implementation is simple with existing infrastructure, but failure detection cannot meet the 100 millisecond requirement for high reliability storage networks
Solution Approach 1:
The patent introduces an intermediary mechanism - the error communication logical port - that mediates between the data communication logical port and other storage nodes. This intermediary actively monitors QP state and communicates error conditions to partner nodes, enabling faster failure detection without completely replacing the existing HCA timeout infrastructure.
3Device complexity
If QP error state is not notified to communication partner, then the erroring QP can remain in error state without active notification, but the normal QP detects partner error only after timeout occurring
Solution Approach 1:
The patent implements feedback by establishing a notification loop where the error communication logical port continuously monitors QP state and provides feedback to partner nodes when errors occur. This feedback mechanism ensures that error states are immediately visible to communication partners, enabling them to take corrective action without waiting for timeout-based detection.
Data Source
AI summary
A storage device according to an embodiment of the present invention has a plurality of storage nodes, each of which has a plurality of logical ports having send and receive queues for a communication request and an identification number, and an internal network for connecting the plurality of storage nodes with one another. The storage nodes each have, as the logical ports, a data communication logical port used for data communication with other storage nodes and an error communication logical port used to notify the other storage nodes of a state of the data communication logical port. When detecting an occurrence of transition of the data communication logical port to an error state, the storage node uses the error communication logical port to notify the other storage nodes of the identification number and the state of the data communication logical port.