Copyless NUMA Hypervisor Memory Migration via Direct Page Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In virtual machines running on NUMA-based architectures, existing memory migration techniques require unnecessary data copying from persistent storage back into the original memory node before migrating to the destination node, leading to computational inefficiencies and increased CPU overhead.
Innovation Solution
A system where a hypervisor exports a shared free list of pages moved to a new source location, allowing virtual machines to detect and map or migrate data directly from this new location to the destination node, eliminating the need for copying back to the original source node.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing memory migration techniques are used in NUMA-based virtual machines, then data can be migrated from source node to destination node, but unnecessary data copying from persistent storage back into the original memory node increases CPU overhead and reduces processing efficiency
Solution Approach 1:
The patent extracts and eliminates the unnecessary intermediate copying step from the memory migration process. By directly migrating data from persistent storage to the destination memory node, the system removes the redundant data copy back to the original source node, thereby reducing CPU overhead and improving processing efficiency.
Solution Approach 2:
The system performs preliminary detection to identify pages that have been migrated to persistent storage before initiating the migration process. This preliminary action enables the system to bypass the original source node and directly access pages from persistent storage, avoiding unnecessary copying operations and reducing CPU overhead.
2Productivity
If existing memory migration techniques are used, then memory pages can be moved between nodes, but the process requires copying data back to the original source node before migrating, leading to computational inefficiencies
Solution Approach 1:
The patent removes the redundant intermediate copying operation from the migration workflow. By extracting this unnecessary step and implementing direct migration from persistent storage to destination node, the system significantly reduces the total time required for memory page migration while improving computational efficiency.
Solution Approach 2:
The system skips the intermediate step of copying data back to the original source node by directly accessing pages from persistent storage. This skipping mechanism allows the migration process to rush through the essential operation without unnecessary delays, thereby reducing migration time and improving computational efficiency.
3Ease of operation
If traditional NUMA balancing operations are performed in virtual machines, then memory balancing can be achieved, but unnecessary data copying increases the complexity of the migration process
Solution Approach 1:
The patent extracts and eliminates the complex intermediate copying operation from the NUMA balancing migration process. By removing this redundant step and implementing direct migration from persistent storage, the system simplifies the overall migration process while maintaining the ability to achieve memory balancing in virtual machines.
Data Source
AI summary
Systems and method for automated efficient memory migration in Non-uniform Memory Access (NUMA) based virtual machines are introduced, comprising exporting one or more informative data objects, from a host machine to at least one virtual machine, wherein the informative data objects include references to pages that have been migrated from a physical source memory node (source node) on the host machine to a new source location on the host machine; inspecting, the informative data objects, by the at least one virtual machine; detecting, by the at least one virtual machine, the pages that have been migrated to the new source location; sending a request, by the at least one virtual machine, to the host machine via the hypervisor, to map the pages to a physical destination memory node (destination node) corresponding to a desired virtual destination memory node (destination vNode); and undertaking at least one efficient data migration operation.