NVMe-oF Virtual Output Queues for Switch Congestion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network attached storage solutions face challenges in providing efficient and reliable storage capabilities, especially when integrating NVMe storage into network switches, due to limitations in handling lossless communication flows and managing congestion, which can lead to data loss or corruption.
Innovation Solution
Integrating SSDs directly into network switches, utilizing virtual output queues and rate control mechanisms to map multiple ingress inputs to a single egress output, and implementing direct feedback control loops to ensure lossless communication and high availability, even in congested conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If NVMe-oF traffic is transmitted over shared network fabric with other traffic, then network resource utilization is improved, but data loss or corruption may occur due to congestion
Solution Approach 1:
The patent segments NVMe-oF traffic into separate virtual output queues (VOQs) within the network switch, isolating storage traffic from other network traffic. This segmentation allows independent management and prioritization of NVMe-oF packets, ensuring data integrity while maintaining high network utilization through dedicated lossless queues.
Solution Approach 2:
The patent introduces virtual output queues as an intermediary mechanism between the network fabric and NVMe storage devices. These VOQs act as buffers that decouple the shared network fabric from the lossless storage traffic requirements, allowing congestion management without data loss.
2Speed
If SSDs are integrated directly into network switches, then network hops are reduced improving speed, but device complexity increases
Solution Approach 1:
The patent merges SSD storage functionality directly into the network switch architecture, creating an integrated system. This consolidation eliminates separate storage devices and network hops, reducing latency and improving data transfer speed while the switch handles both networking and storage functions.
Solution Approach 2:
The network switch is designed with multi-functionality, serving both as a network routing device and as a storage system. By integrating NVMe-oF capabilities into the switch fabric, the device performs dual roles, reducing infrastructure complexity despite increasing individual device functionality.
3Productivity
If multiple ingress inputs are mapped to a single egress output, then storage access efficiency is improved, but congestion management becomes more difficult
Solution Approach 1:
The patent segments the egress output into multiple virtual output queues, each handling traffic from different ingress inputs. This segmentation allows independent congestion control and rate limiting for each queue, simplifying traffic management while maintaining high storage access efficiency through parallel processing paths.
Solution Approach 2:
The patent implements feedback mechanisms that monitor queue depths and congestion levels in real-time. This feedback enables dynamic rate adjustment and traffic shaping, allowing efficient handling of multiple ingress inputs to a single egress output while preventing congestion through adaptive control.
Data Source
AI summary
A network infrastructure device (e.g., network switch), that integrates solid-state drive (SSD) storage, using Non-volatile Memory Express (NVMe) data transfer protocol, for use by remote application hosts is provided. High availability configurations of network switches using direct rate control (RC) feedback for a plurality of submission queues mapped to SSD storage is provided. NVMe over fabric (NVMe-oF) is an implementation of NVMe protocol over a network fabric. Access to SSDs over network fabrics may be controlled using a direct RC feedback signal between an egress queue congestion accounting (associated with a single egress output) and a source node receiving input/output commands from remote hosts for the integrated SSD devices. In some implementations, direct RC feedback signals use hardware based signals. In some implementations, direct RC feedback signals are implemented in the hardware logic (silicon chip logic) within an internal switch fabric of the network switch.