Host-Based NVM Clustering via Network Mapped Queues

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for utilizing non-volatile memory (NVM) in enterprise network storage fail to achieve desired access specifications, including high speeds, high I/O operations per second, low interconnect latency, and seamless data transport, due to limitations in interacting with network elements to utilize full performance potential.

Innovation Solution

A method involving the initiation of cluster parameters for NVM clusters, mapping submission and completion queues based on shared NVM details across nodes, using a first-in-first-out scheme, and multicasting cluster parameters to enable seamless, high-performance transport and clustering of host-based SSDs with central storage appliances.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If current methods of utilizing NVM are used, then NVM can store data without power, but it cannot achieve desired access speeds similar to DDR3 SDRAM

Engineering Contradiction:

Improvedata access speedVSAvoidnetwork infrastructure complexity

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent introduces a network mapping mechanism that acts as an intermediary between NVM devices and the network infrastructure. Submission and completion queues are mapped to network addresses, enabling direct memory access over the network without complex protocol stacks, thus achieving DDR3-like speeds while maintaining network connectivity.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional mechanical storage access methods with a memory-mapped I/O approach. By mapping NVM submission and completion queues to network addresses, the system eliminates the need for complex network protocol processing, achieving speeds comparable to local memory while using network infrastructure.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If current NVM implementation methods are used, then NVM can retain information without power, but it cannot achieve one million IOPS or more

Engineering Contradiction:

ImproveIOPSVSAvoidinterconnect latency

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent pre-allocates and maps submission and completion queues to network addresses before actual data operations. This preliminary mapping action enables direct memory access without runtime address resolution, reducing latency and enabling the system to achieve one million IOPS or more by eliminating address translation overhead during data operations.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a virtual copy of the NVM address space in the network address space through mapping. This allows the system to access NVM devices as if they were local memory, achieving high IOPS by eliminating network protocol overhead while maintaining the durability benefits of NVM.

Inventive Principle:

Principle #26Copying

3Loss of time

If current NVM implementation methods are used, then NVM can store data persistently, but it cannot achieve interconnect latency between 3 μs and 10 μs

Engineering Contradiction:

Improveinterconnect latencyVSAvoidclustering mechanism complexity

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The patent merges the NVM cluster management functionality with the existing network file system infrastructure. By integrating submission and completion queue mapping into the network file system protocol, the system achieves low interconnect latency without requiring a separate complex clustering mechanism, as the existing network infrastructure handles the coordination.

Inventive Principle:

Principle #5Merging (Combining)

4Ease of operation

If current NVM implementation methods are used, then NVM can provide storage capacity, but it cannot provide seamless high-performance transport of data in the network infrastructure

Engineering Contradiction:

Improvedata transport seamlessVSAvoiddata transport performance

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The patent makes the NVM submission and completion queues accessible through universal network protocols. By mapping these queues to network addresses that can be accessed through standard network file system operations, the system provides seamless high-performance data transport that works with existing network infrastructure without requiring specialized protocols or interfaces.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11321271B2Host based non-volatile memory clustering mechanism using network mapped storage

Publication Date: 2022.05.03 KYNDRYL INC

AI summary

In one embodiment, a method includes initiating cluster parameters that govern how a non-volatile memory (NVM) cluster functions and operates. Submission and completion queues of shared NVM on other nodes in the NVM cluster are mapped based on details of the shared NVM on the other nodes in the NVM cluster. The submission queue is configured to store commands to access the shared NVM according to a first-in-first-out (FIFO) scheme. The completion queue is configured to store completed commands after being processed through the submission queue. In another embodiment, a host-based data storage system includes NVM configured to store data. The host-based data storage system further includes a processor and logic integrated with and/or executable by the processor to perform the foregoing method.