Hardware Accelerator VF Workload Handoff to Reduce Dirty Page Copying
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The transfer of memory pages during VM live migration with hardware accelerators, such as GPUs, results in excessive network bandwidth usage and failure to meet service-level agreements due to the large number of dirty pages, leading to potential downtime and reduced success rates.
Innovation Solution
A hardware status manager identifies workloads targeting a hardware accelerator VF and determines whether to execute them on the source or destination VM, bypassing the transfer of dirty pages by executing the workload on the destination VM, thereby reducing the need for network bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If memory pages are transferred during live migration with hardware accelerators, then the VM migration can be completed, but network bandwidth usage becomes excessive and service-level agreements are not met
Solution Approach 1:
The patent extracts and identifies dirty pages generated by hardware accelerator workloads separately from regular memory pages. By detecting which pages are modified by GPU/VF operations and excluding them from the migration transfer, the system removes the harmful factor (excessive bandwidth consumption) while preserving the essential migration function.
Solution Approach 2:
The patent applies different handling strategies to different types of memory pages based on their origin. Regular pages are migrated normally, while pages dirtied by hardware accelerator workloads are identified and excluded from migration. This localized differentiation resolves the contradiction by treating only the problematic subset of pages differently.
2Reliability
If all dirty pages are transferred during live migration, then complete VM state migration is achieved, but migration time increases and downtime is extended
Solution Approach 1:
The patent extracts dirty pages generated by hardware accelerator workloads from the overall dirty page set and excludes them from migration transfer. This extraction reduces the total volume of data to be transferred, directly decreasing migration time and downtime while maintaining migration success through selective page handling.
Solution Approach 2:
The patent performs preliminary identification of hardware accelerator workload pages before the migration transfer begins. By pre-detecting which pages will be modified by accelerators and excluding them in advance, the system avoids unnecessary transfer operations, reducing both migration time and downtime.
3Stability of the object's composition
If dirty pages from hardware accelerator workloads are transferred, then memory consistency is maintained, but network bandwidth is excessively consumed
Solution Approach 1:
The patent extracts and identifies pages modified by hardware accelerator workloads and excludes them from migration. This removal maintains memory consistency for the transferred pages while eliminating the bandwidth consumption caused by transferring accelerator-generated dirty pages that would be regenerated anyway on the destination.
Solution Approach 2:
The patent applies different migration strategies to different page types: regular pages are transferred to maintain consistency, while hardware accelerator pages are excluded. This local differentiation resolves the contradiction by maintaining consistency only where necessary while saving bandwidth on pages that will be regenerated.
Data Source
AI summary
Embodiments described herein are generally directed to an improved workload submission handling strategy for workloads received during live migration and targeting VFs of a hardware accelerator. In an example, while performing a live migration of a source VM running on a source host to a destination VM of a destination host, a new workload targeting a VF of a first HW accelerator of the source host is identified by a HW status manager of the source host by trapping the workload submission channel. Based on a nature of the new workload, the HW status manager determines whether to transfer the new workload to the destination host. Responsive to an affirmative determination, the HW status manager causes the new workload to be submitted to a VF of a second HW accelerator of the destination host by incorporating information regarding the new workload within a migration stream associated with the live migration.