Network-Attached Memory for Virtual Machine Migration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual machine migration techniques result in significant downtime due to repeated memory copying and the need for checkpointing, which is unacceptable in busy workloads, as the process is slowed by increasing 'dirty' memory pages during migration.
Innovation Solution
Implementing a network-attached memory system that allows virtual machines to run on either server with memory residing on a high-speed network-attached memory device, enabling instantaneous migration without memory copying by moving memory from one server to another while the workload continues to run, thus eliminating downtime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If repeated memory copying is performed during virtual machine migration, then memory consistency is maintained, but migration time and downtime increase
Solution Approach 1:
The system performs preliminary memory copying operations during the migration process, copying memory pages before they become dirty. This preliminary action reduces the need for repeated copying operations and minimizes the final checkpoint duration, thereby maintaining memory consistency while reducing overall migration time
Solution Approach 2:
A third copy location is introduced as an intermediary during the migration process. Memory is copied to this intermediate location while the virtual machine continues to run, allowing the source and destination systems to be decoupled. This intermediary enables parallel copying operations and reduces the impact on migration time while maintaining data consistency
2Reliability
If the virtual machine is frozen for final memory copy, then data consistency is achieved, but service availability decreases
Solution Approach 1:
Instead of freezing the virtual machine for the entire migration process, the system only freezes it partially and briefly for the final memory copy operation. The majority of the copying is performed while the virtual machine continues to run, achieving data consistency with minimal service interruption
Solution Approach 2:
The virtual machine continues to run and process requests throughout most of the migration process. The freezing action is minimized to only the final checkpoint, maintaining continuous useful action and service availability while still achieving the necessary data consistency for successful migration
3Ease of operation
If memory copying is performed while workload is busy, then migration can proceed online, but the number of dirty memory pages increases
Solution Approach 1:
The system performs preliminary copying of memory pages that are likely to be accessed soon, before they become dirty during normal operation. This preliminary action reduces the accumulation of dirty pages during online migration, enabling continuous operation while managing memory copy requirements
Solution Approach 2:
An intermediate copy location serves as a buffer, allowing memory to be copied to this location while the virtual machine continues to access and modify the original memory. This intermediary enables online migration by decoupling the copying process from the execution process, managing dirty page accumulation without requiring system freezes
Data Source
AI summary
A network system comprising a plurality of servers communicatively-coupled on a network, a network-attached memory coupled between a first server and a second server of the server plurality, and a memory management logic that executes on selected servers of the server plurality and migrates a virtual machine from the first server to the second server with memory for the virtual machine residing on the network-attached memory.