Kernel I/O Thread for Virtual Storage Request Processing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In virtualization scenarios, the processing of read/write requests is time-consuming due to the complexity of the storage software stack, leading to significant delays, especially with high-speed storage media where the storage software stack becomes a bottleneck, with delays reaching 19.3% for SSDs and 94.1% for DDR NVM media.

Innovation Solution

A method and apparatus that actively poll the instruction transmit queue of a virtual storage device to obtain read/write requests, forward them to a physical storage device, and process them without needing virtual machine exit and enter operations, using memory-mapped I/O to simplify the process and reduce delays.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a conventional storage software stack is used in virtualization scenarios, then storage services can be provided to virtual machines, but processing delays increase significantly (up to 94.1% for DDR NVM media)

Engineering Contradiction:

Improvestorage service availabilityVSAvoidrequest processing delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts the I/O processing function from the complex virtualization software stack (QEMU, KVM, virtio) and implements it directly in the kernel through a dedicated I/O processing thread. This removes multiple software layers that cause delays, while maintaining storage service availability for virtual machines.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the I/O processing into distinct components: a kernel-space I/O processing thread handling request forwarding, a separate mapping mechanism for virtual-to-physical device translation, and an interrupt handling subsystem. This segmentation allows each component to operate efficiently independently, reducing overall processing delay.

Inventive Principle:

Principle #1Segmentation

2Ease of operation

If QEMU process wake-up mechanism is used for I/O handling, then virtual machine I/O requests can be processed, but processing speed decreases due to context switching overhead

Engineering Contradiction:

ImproveI/O request handling capabilityVSAvoidrequest processing speed

Core Design Contradiction:

Ease of operationVSSpeed

Solution Approach 1:

The patent replaces the mechanical QEMU process wake-up mechanism with a kernel-space I/O processing thread that continuously monitors and handles I/O requests. This substitution eliminates context switching overhead and process wake-up delays, significantly improving request processing speed while maintaining full I/O handling capability.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If virtual machine exit and enter operations are required for I/O processing, then storage virtualization can be implemented, but processing efficiency decreases due to additional operation overhead

Engineering Contradiction:

Improvestorage virtualization capabilityVSAvoidI/O processing throughput

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent introduces a kernel-space I/O processing thread as an intermediary between virtual machine I/O requests and the storage subsystem. This intermediary handles virtual-to-physical device mapping and request forwarding without requiring virtual machine exit/enter operations, maintaining storage virtualization capability while dramatically improving processing throughput.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Adaptability or versatility

If a complex virtualized storage software stack is used, then storage services can be provided to multiple virtual machines, but device complexity increases

Engineering Contradiction:

Improvemulti-virtual machine supportVSAvoidsoftware stack complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements a universal kernel-space I/O processing thread that handles I/O requests from multiple virtual machines through a unified interface. This single thread provides multi-functionality by managing virtual-to-physical device mapping, request forwarding, and interrupt handling for all virtual machines, reducing software stack complexity while maintaining multi-VM support.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS10579305B2Method and apparatus for processing read/write request in physical machine

Publication Date: 2020.03.03 HUAWEI TECH CO LTD

AI summary

A method and an apparatus for processing a read/write request in a physical machine, where the method includes polling, by a host by accessing memory of at least one of virtual storage devices, at least one instruction transmit queue of the at least one virtual storage device in order to obtain a first read/write request from the at least one instruction transmit queue, performing a first forwarding operation on the first read/write request, and obtaining, by the host, another first read/write request from the at least one instruction transmit queue by polling such that the host performs the first forwarding operation on the other first read/write request. According to the method and the apparatus in embodiments of the present disclosure, a speed of processing a read/write request in a virtualization storage scenario can be increased.