Container Live Migration via DMA and RDMA CPU Offload
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing container live migration methods consume excessive resources and slow down the CPU, disrupting other services on the host during the migration process.
Innovation Solution
A processor offloads the data transmission process from the CPU by using direct memory access (DMA) to transmit container data between hosts, utilizing a private network for accelerated data transfer and incorporating RDMA for direct memory access to the target host, thereby reducing CPU resource consumption and improving migration efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the CPU transmits all container data from source host to target host, then container live migration is achieved, but CPU resource consumption becomes excessive
Solution Approach 1:
A dedicated processor is introduced as an intermediary between the CPU and the data transmission task. This processor specifically handles the data copying and network transmission operations, while the CPU focuses on container management and orchestration. The intermediary processor eliminates the burden of data transmission from the CPU, resolving the contradiction between migration capability and CPU resource consumption.
Solution Approach 2:
The data transmission function is extracted from the CPU and assigned to a separate dedicated processor. By taking out the data copying and network transmission tasks from the CPU's workload, the system allows the CPU to concentrate on higher-level container management functions, thereby reducing CPU resource consumption while maintaining migration productivity.
2Productivity
If the CPU transmits container data during migration, then migration is completed, but other services on the host are disrupted
Solution Approach 1:
The dedicated processor acts as an intermediary that isolates the data transmission operations from the CPU and other services. By having a separate processor handle the resource-intensive data copying and network transmission, the CPU can continue managing other services without interruption, thus reducing the harmful impact of migration on other host services.
Solution Approach 2:
The migration process is segmented into distinct functions handled by different processors. The dedicated processor handles data transmission independently, while the CPU continues managing container orchestration and other services. This segmentation allows migration operations to proceed without blocking or disrupting other service operations on the host.
3Productivity
If traditional CPU-based data transmission is used, then migration is achieved, but migration speed is slow
Solution Approach 1:
The dedicated processor serves as an intermediary optimized specifically for data transmission tasks. It incorporates dedicated data copying mechanisms and network interfaces that are more efficient than general-purpose CPU-based transmission. This intermediary processor can perform bulk data copying at higher speeds, thereby accelerating the overall migration process.
Solution Approach 2:
The system replaces the mechanical CPU-based data transmission mechanism with a dedicated processor that uses optimized hardware-level data copying and network transmission. This substitution eliminates the software-based processing overhead associated with CPU data transmission, resulting in faster data transfer speeds and accelerated container migration.
Data Source
AI summary
This application discloses a container live migration method, a processor, a host, a chip, and an interface card. In this application, because data of a container is obtained from a memory of a source host through DMA, and the data of the container is transmitted to a target host instead of using a CPU, the CPU of the source host does not need to perform, in a live migration process, an operation of transmitting the data of the container.