RDMA Virtual Machine State Recovery

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In virtualized computing environments, existing system recovery methods require significant CPU resources and double the system requirements by necessitating a standby computer for each working virtual machine, making them inefficient and costly.

Innovation Solution

The method employs remote direct memory access (RDMA) over InfiniBand/RoCE to transfer the state of a virtual computer system from a physical computer system to a standby system, allowing for fast recovery without the need for multiple standby systems, utilizing RDMA operations to pull the state records and re-initialize I/O systems through non-volatile RAM.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If a standby computer is kept for each working virtual machine to enable fast recovery, then recovery speed is improved, but system resource requirements double

Engineering Contradiction:
Improverecovery speedVSAvoidsystem resource requirements
Core Design Contradiction:
SpeedVSQuantity of substance

Solution Approach 1:

The patent merges multiple standby functions into a single physical server by virtualizing the standby environment. Instead of requiring one standby computer per working virtual machine, the system creates a consolidated standby server that can host multiple standby virtual machines, thereby reducing hardware requirements while maintaining fast recovery capability

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The standby server is designed with multi-functionality to serve multiple working virtual machines simultaneously. A single standby server can assume the role of any failed working virtual machine through dynamic virtual machine migration and state restoration, making the standby infrastructure universally applicable across different virtual machine failures

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If backup agents are deployed on the operating system for each application to enable data backup, then data protection is improved, but CPU resource consumption increases significantly

Engineering Contradiction:
Improvedata protectionVSAvoidCPU resource consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The backup functionality is extracted from the operating system level and moved to the virtualization platform level. Instead of deploying backup agents on each application or OS, the patent implements backup at the virtual machine image level, which consolidates backup operations and eliminates the need for multiple individual backup agents, thereby reducing CPU overhead

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system creates copies of virtual machine images and their associated file systems at the virtualization layer for backup purposes. This approach allows complete system state replication without requiring active backup agents during operation, enabling fast recovery through image restoration while minimizing impact on running system performance

Inventive Principle:
Principle #26Copying

Data Source

PatentUS10061744B2Using RDMA for fast system recovery in virtualized environments
Publication Date: 2018.08.28 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • US10061744B2 patent drawing
  • US10061744B2 patent drawing
  • US10061744B2 patent drawing

AI summary

A method, system and computer program product are disclosed for recovery in a virtualized environment using remote direct memory access (RDMA). In one embodiment, the method comprises operating a virtual computer system on a physical computer system, and the virtual system maintains in a memory area a record of a state of the virtual system. In this method, when defined error conditions occur on the virtual system, RDMA is used to pull the record of the state of the virtual system from that memory area onto a standby computer. This record on the standby computer is used to re-initialize the virtual computer. Embodiments of the invention provide methods that provide a very fast recovery from a virtual machine fault or error, while requiring much fewer resources than standard approaches. In embodiments of the invention, one spare real computer system can be used for backing up several virtual systems.