SMB PMEM Dialect for Remote Non-Volatile Memory Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current solutions do not allow direct access to non-volatile memory over a network using protocols like SMB or NFS, limiting remote access and efficient data transfer between servers, especially for OS deployment and storage operations.
Innovation Solution
A multiple processor system utilizing an inter-processor messaging protocol to enable remote direct memory access to non-volatile memory devices, reducing dependency on local file systems and increasing throughput by using RDMA queues and SMB PMEM dialect for direct stack transactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If non-volatile memory is accessed through a local processor using a file system, then the memory can be accessed reliably, but the access speed is limited and remote access is not possible
Solution Approach 1:
The patent introduces an inter-processor messaging protocol as an intermediary mechanism that enables remote processors to access non-volatile memory without requiring traditional file system operations. The protocol acts as a mediator between the remote processor and the memory device, allowing direct memory access while maintaining reliability.
Solution Approach 2:
The patent extracts the file system dependency from the memory access path by implementing direct memory access through RDMA queues. This removes the intermediate file system layer that limited access speed and remote capability, allowing processors to access memory directly at hardware level speeds.
2Productivity
If traditional RDMA is used for remote memory access, then network operations can be CPU-independent, but it requires a PMEM aware file system that increases complexity
Solution Approach 1:
The patent extracts the PMEM aware file system requirement from the system architecture by implementing direct access to non-volatile memory through RDMA queues. This eliminates the need for complex file system software layers while maintaining CPU-independent network and storage operations.
Solution Approach 2:
The patent makes RDMA queues universally compatible with non-volatile memory devices by implementing a standardized interface that works with different memory types and configurations. This universal approach allows CPU-independent operations without requiring specialized file systems for each memory type.
3Speed
If direct access to non-volatile memory is implemented without a local processor, then access speed increases, but reliability during OS crashes or pre-OS modes becomes problematic
Solution Approach 1:
The patent implements beforehand cushioning by establishing RDMA queue connections and memory mappings before OS crashes or failures can occur. The system prepares the direct access pathway in advance during normal operation, so that even if the OS crashes or enters pre-OS mode, the hardware-level RDMA connection remains intact and reliable.
Solution Approach 2:
The RDMA protocol acts as a reliable intermediary that maintains connection stability during OS failures. The hardware-level RDMA infrastructure continues to function independently of software state, providing consistent access to non-volatile memory even when the operating system is unavailable or crashed.
4Reliability
If file system operations are used for network and storage I/O, then data can be managed reliably, but transaction overhead increases and throughput decreases
Solution Approach 1:
The patent extracts file system operations from the I/O path by implementing direct memory access through RDMA queues. This removes the software processing overhead of file system operations while maintaining data integrity through hardware-level transaction support in the RDMA protocol.
Solution Approach 2:
The patent replaces the mechanical file system processing layer with a more efficient hardware-level RDMA mechanism. The RDMA queues provide direct memory access at the hardware level, substituting the software-based file system operations with a faster, hardware-optimized pathway that maintains reliability through built-in error checking and transaction support.
Data Source
AI summary
A multiple processor system comprising a first processor configured to utilize an inter-processor messaging protocol and a second processor configured to utilize the inter-processor messaging protocol is disclosed. The system includes a non-volatile memory device coupled to the first processor, wherein the second processor is configured to access the non-volatile memory device by utilizing the inter-processor messaging protocol.


