NVMeoF Host I/O Request Path Selection via Access Interval Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In NVMe over Fabrics (NVMeoF) storage systems, the forwarding of I/O requests between control nodes increases I/O delay due to the need for calculation and path selection based on logical addresses, which is inefficient and can lead to increased latency.
Innovation Solution
Implementing an access interval allocation method for each control node, where the host sets and manages access intervals for control nodes, allowing for direct delivery of I/O requests to the appropriate node based on logical address mapping, thereby reducing the need for path forwarding and minimizing delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If I/O requests are forwarded between control nodes based on calculation of logical addresses, then the storage system can achieve flexible path selection and load distribution, but the I/O delay increases due to the forwarding process
Solution Approach 1:
The host performs preliminary calculation of the target control node based on the logical address before sending the I/O request. This preliminary action eliminates the need for intermediate forwarding nodes to perform calculation, thereby reducing I/O delay while maintaining flexible path selection capability
Solution Approach 2:
The calculation function is extracted from the control nodes and moved to the host. The host calculates which control node should execute the I/O request based on the logical address, and directly forwards the request to that node without requiring intermediate forwarding calculations, thus reducing processing delay
2Reliability
If multiple control nodes are used to access logical disks, then system reliability and load distribution are improved, but device complexity increases due to path management
Solution Approach 1:
The host autonomously performs the calculation to determine the target control node based on the logical address and a preset algorithm. This self-service approach eliminates the need for complex centralized path management mechanisms, reducing system complexity while maintaining multi-node access capability for improved reliability
3Loss of time
If I/O requests are directly delivered to the target control node without forwarding, then I/O delay is reduced, but the host must perform additional calculation based on logical addresses
Solution Approach 1:
The host is designed with universal capability to perform the calculation algorithm for determining target control nodes. By making the host multi-functional (capable of both I/O processing and control node selection calculation), the system eliminates intermediate forwarding steps, reducing I/O delay while distributing the computational burden to the host
Data Source
Figure 1
Figure 2~3a
Figure 3b
AI summary
Embodiments of the present invention provide a data processing method performed by a host. The host and a storage system communicate with each other by using the NVMeoF protocol. The storage system includes a logical disk, and the host accesses the logical disk by using a control node in the storage system. The method includes: sending a state query command to the control node, where the state query command is used to instruct the control node to report a path state of a path corresponding to the control node; receiving the path state reported by the control node; when the received path state indicates that the logical disk includes an access interval, sending an interval query command to the control node; receiving access interval information reported by the control node, where an access interval indicated by the access interval information is pre-allocated to the control node; and recording a mapping relationship between the control node and the access interval information.