NVMe over Network Protocol for Ethernet Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current solutions for scale-out applications face challenges with inconsistent performance, high infrastructure costs, and complex data transfer processes due to the proliferation of IO operations across multiple software stacks and networks, leading to inefficiencies and poor performance in hypervisor and bare metal solutions.
Innovation Solution
The implementation of a protocol, NVMEoN, which enables guaranteed delivery of NVMe commands and data over Ethernet networks with minimal latency, using a Burst Transmission Protocol and a converged storage and network controller to reduce IO operations and improve scalability, allowing for dynamic resource allocation and high-performance multi-tenant deployments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transferred across multiple software stacks and networks in hypervisor or bare metal solutions, then data transfer can be achieved, but IO operations increase and performance becomes inconsistent
Solution Approach 1:
The patent combines storage and network controllers into a single converged controller that handles both storage protocols (NVMe, Fibre Channel) and network protocols (Ethernet, InfiniBand) through a unified architecture. This merging eliminates the need for separate software stacks for storage and network operations, reducing IO operation complexity while maintaining high data transfer efficiency through direct hardware paths.
Solution Approach 2:
The converged controller acts as an intermediary between the host system and multiple storage/network protocols. It translates and routes commands between different protocols without requiring multiple software stacks, thereby simplifying the IO path while enabling efficient data transfer across different network and storage mediums.
2Adaptability or versatility
If conventional storage protocols are used over Ethernet networks, then network compatibility is improved, but latency increases due to protocol overhead
Solution Approach 1:
The patent segments the storage and network protocol handling into separate functional modules within the converged controller. NVMe commands are processed separately from Ethernet networking functions, allowing optimized paths for each protocol type. This segmentation enables the system to maintain low latency for NVMe operations while simultaneously supporting multiple network protocols for compatibility.
Solution Approach 2:
The converged controller provides different optimization levels for different protocols handled by the same device. NVMe commands receive priority handling with minimal overhead, while other network protocols operate with standard processing. This local quality approach ensures high performance for time-sensitive storage operations while maintaining broad network compatibility.
3Speed
If NVMe is accessed via PCIe interface directly, then latency is minimized, but network accessibility is lost
Solution Approach 1:
The converged controller provides universal access to NVMe storage devices through multiple interfaces including PCIe, Ethernet, and InfiniBand. The controller maintains the low-latency PCIe path for direct-attached storage while simultaneously enabling network accessibility through protocol translation and remote direct memory access (RDMA) capabilities, making the same storage device accessible both locally and remotely with minimal performance degradation.
Data Source
AI summary
Enabling a protocol for efficiently and reliably using the NVME protocol over a network, referred to as NVME over Network, or NVMEoN, may include an NVMEoN exchange layer for handling exchanges between initiating and target nodes on a network, a burst transmission protocol that provides guaranteed delivery without duplicate retransmission, and an exchange status block approach to manage state information about exchanges.