Virtual Machine Suspend Resume Metadata Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Virtual machine (VM) suspend and resume operations in virtualized computer systems often result in prolonged downtime due to the time-consuming process of writing and reading the VM's state to a file, leading to network connection expiration and user disconnection, especially for VMs with large memory sizes.
Innovation Solution
A method that instantiates a second VM instance while the first is running, suspends the original VM, copies metadata for virtual memory pages instead of memory, and loads the stored state into the new VM, allowing for configuration changes without noticeable downtime or performance degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the VM's state is written to a file and then read back, then the VM configuration can be changed, but the downtime becomes very long (tens of seconds to minutes)
Solution Approach 1:
The invention extracts only the essential metadata (page tables, device states, register values) from the complete VM state, excluding the large memory contents. This selective extraction reduces the serialization size from gigabytes to kilobytes or small megabytes, enabling rapid save/restore operations that minimize downtime while preserving configuration change capability
Solution Approach 2:
The VM state is segmented into two distinct parts: metadata (control structures, page tables, device states) and memory contents. Only the metadata portion is serialized to disk during suspend operations, while memory is handled separately through pre-copying or direct mapping between old and new VM instances. This segmentation allows configuration changes to be applied to the metadata without requiring full memory serialization
2Loss of time
If a new VM instance is created with pre-copying of memory, then downtime is reduced, but CPU and memory usage double
Solution Approach 1:
Instead of performing a complete memory pre-copy operation, the invention applies partial action by only copying or serializing the metadata structures (page tables, device states, registers). The actual memory contents are either shared between old and new VM instances or handled through efficient direct mapping, reducing the copying overhead from 100% of memory to a small fraction (metadata only), thus lowering CPU and memory usage while maintaining fast resume performance
Data Source
AI summary
A virtual machine is suspended and quickly restarted while maintaining the VM's state. The method is quick enough so that network connections are maintained across the restart and the guest operating system and guest applications running in the VM are not aware of the restart. As a result, users and clients connected to the VM do not notice any downtime or disruption to the VM. After suspension and before the restart, VM configuration changes that would not be possible or be very difficult through code changes alone while the VM was running can be made.