Live Partition Mobility with Cached Memory Pre-Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current live partition migration technologies are lengthy and disruptive for virtual machines with large memory footprints, impacting performance and response time, especially for active workloads like in-memory databases.
Innovation Solution
Implement a local cache on the destination computer to pre-transfer least recently modified memory pages of the source virtual machine before the actual migration event, allowing parallel building of the destination VM from the cache and direct transfer of remaining pages during the event.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If pre-transfer of memory pages is implemented, then migration speed is improved, but device complexity increases
Solution Approach 1:
The system performs preliminary actions by pre-transferring least recently modified (LRU) memory pages from the source VM to the destination VM before the actual migration event. This pre-transfer occurs in the background without impacting application performance, and when migration is triggered, the already-transferred pages are immediately available, significantly reducing migration time and accelerating the overall process.
Solution Approach 2:
The memory transfer process is segmented into two distinct phases: a pre-transfer phase that occurs in the background before migration, and a completion phase that occurs during the actual migration event. This segmentation allows the system to prepare data in advance without blocking application operations, and then quickly finish the migration using the pre-prepared data, thus improving speed while managing complexity through phased execution.
2Loss of time
If continuous pre-transfer of memory pages is initiated, then application response time is improved, but network bandwidth consumption increases
Solution Approach 1:
The system applies local quality by selectively transferring only the least recently modified (LRU) memory pages during the pre-transfer phase, rather than transferring all memory pages uniformly. This targeted approach minimizes network bandwidth consumption by focusing only on pages that are least likely to be accessed, while still achieving the goal of reducing application response time during migration.
Solution Approach 2:
The system performs partial action by transferring only a subset of memory pages (specifically the LRU pages) in advance, rather than completing the full transfer before migration. This partial pre-transfer is sufficient to significantly reduce migration impact on application response time, while consuming less network bandwidth than a complete pre-transfer would require.
3Adaptability or versatility
If LPM is performed for VMs with large memory footprint, then migration capability is improved, but migration duration increases
Solution Approach 1:
For VMs with large memory footprints, the system performs preliminary action by initiating background pre-transfer of LRU memory pages before the migration event. This allows the migration process to start with already-transferred pages, significantly reducing the duration of the actual migration event while maintaining the capability to migrate large VMs that would otherwise be too slow to migrate.
Solution Approach 2:
The system maintains continuity of useful action by performing memory page transfers continuously in the background during normal VM operation, rather than interrupting the VM for the transfer. This continuous background operation prepares data for future migration without impacting the VM's current performance, enabling large VMs to be migrated faster when needed while maintaining migration capability.
Data Source
AI summary
A method, computer system, and computer program product are provided for non-disruptive live partition mobility (LPM). An orchestrator verifies source profile of virtual machine (VM) indicates source VM instant live partition mobility (LPM) capable. If so, orchestrator instructs source computer hypervisor to initiate handshake with destination computer hypervisor. When instant LPM capable VM is activated, source computer hypervisor enables source virtual input/output server (VIOS) to initiate continuous pre-transfer of source VM memory pages to virtual input/output server (VIOS) on destination computer. Received source VM memory pages are written to local cache on destination computer. Pre-transferred source VM memory pages are least recently modified memory pages. Upon initiating LPM event the hypervisor receives an instruction to build a destination VM. In parallel to build of destination VM, remaining source VM memory pages are sent to the destination computer. Upon completion of the build of the destination VM, terminating the LPM event.