Smart NIC Synchronous Data Access via RDMA Metadata Offload
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies lack a synchronous interface for persistent memory, leading to significant CPU overhead in data access operations, which hampers the application of persistent memory in storage systems and data protection systems.
Innovation Solution
A method involving a programmable network device with RDMA functions is used to send metadata for data movement, allowing the device to perform data operations while the CPU enters a hibernation state, and exits upon confirmation of operation completion, thereby reducing CPU workload and enabling synchronous data access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If persistent memory is accessed in DAX mode, then data access performance is improved, but CPU overhead increases significantly due to memory copying operations
Solution Approach 1:
A programmable network device with RDMA functions is introduced as an intermediary between the CPU and persistent memory. This device handles data movement and access operations independently, allowing the CPU to enter hibernation state during data transfer and wake up only when operations complete, thus reducing CPU overhead while maintaining fast data access performance
Solution Approach 2:
The programmable network device performs data movement and access operations autonomously without continuous CPU intervention. The device can execute data transfer tasks independently and notify the CPU when operations complete, enabling the CPU to minimize its involvement and reduce overhead while still achieving synchronous data access
2Reliability
If synchronous data access is implemented, then data protection reliability is improved, but CPU resource utilization increases
Solution Approach 1:
The programmable network device acts as a mediator that implements synchronous data access operations independently. It manages data movement and synchronization tasks without requiring continuous CPU resources, thereby maintaining data protection reliability while minimizing CPU utilization through autonomous operation and hibernation capabilities
3Use of energy by moving object
If CPU enters hibernation state during data movement, then CPU overhead is reduced, but system response time may increase
Solution Approach 1:
The programmable network device implements a feedback mechanism that notifies the CPU when data movement operations complete. This allows the CPU to transition between hibernation and active states efficiently, minimizing the impact on system response time while maximizing CPU overhead reduction during data transfer operations
Data Source
AI summary
Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for synchronously accessing data. The method may include sending metadata associated with data to be moved by a user to a programmable network device associated with a persistent memory containing the data, so as to enable the programmable network device to move the data based on the metadata, wherein the programmable network device is a smart network interface card having a remote direct memory access function. The method may also include entering a hibernation state. In addition, the method may include exiting from the hibernation state in response to receiving a confirmation of operation completion from the programmable network device, so as to notify the user that an operation of moving the data is complete. By means of the embodiments of the present disclosure, an operation of synchronously accessing data can be implemented, and computing resources of the CPU are saved, thereby improving user experience.