Virtual Machine Manager Offloading for Fast DPU-Based VM Startup
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Concurrent starting of virtual machines in cloud computing environments is hindered by high latency and resource competition due to weak communication channels and limited CPU capabilities of Data Processing Units (DPU), leading to prolonged VM Exit processes and inability to meet user demands for rapid capacity expansion.
Innovation Solution
Create virtual machine managers on the host with abundant CPU resources to concurrently start virtual machines, and then offload these managers to the DPU for management once the starting process is complete, leveraging the DPU's efficient resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If virtual machine managers are created and started on the DPU, then the DPU can directly manage virtual machines with reduced host involvement, but the weak communication channels and limited CPU capabilities of the DPU cause high latency and resource competition, prolonging the VM Exit process
Solution Approach 1:
The patent segments the virtual machine management process into two distinct phases: a startup phase where virtual machine managers are created and initialize virtual machines on the host, and a runtime phase where managers are offloaded to the DPU. This segmentation allows the time-sensitive startup operations to occur on the host while the DPU handles ongoing management, resolving the contradiction between direct DPU management and startup performance.
Solution Approach 2:
The patent performs preliminary actions by creating virtual machine managers on the host before offloading them to the DPU. The host prepares and initializes the managers with necessary configurations and resources, so that when the DPU takes over, the managers are already optimized and ready for efficient operation, reducing the VM Exit process duration.
2Productivity
If multiple virtual machines are started concurrently on the host, then rapid capacity expansion can be achieved, but resource competition on the host increases and load pressure rises
Solution Approach 1:
The patent extracts the virtual machine manager workload from the host and transfers it to the DPU after the startup phase. This extraction removes the ongoing management burden from the host, allowing the host to handle concurrent startup requests without excessive load pressure while the DPU manages the running virtual machines.
Solution Approach 2:
The patent implements a dynamic workload distribution model where the location of virtual machine managers transitions from host to DPU based on the operational phase. During startup, managers reside on the host for rapid provisioning; after offloading, they operate on the DPU for efficient runtime management. This dynamic approach optimizes both startup speed and host load distribution.
3Productivity
If the DPU handles virtual machine management directly, then resource utilization on the DPU is improved, but the communication latency between host and DPU increases and resource competition occurs
Solution Approach 1:
The patent performs preliminary configuration and initialization of virtual machine managers on the host before offloading them to the DPU. This preliminary action ensures that all necessary parameters, resource allocations, and management policies are pre-configured, minimizing the need for communication between host and DPU during runtime and reducing communication latency.
Data Source
AI summary
The embodiments of the present disclosure provide a method and a device for controlling a starting of a virtual machine based on a data processing unit (DPU). The method comprises: sending, in response to receiving an instruction for starting a plurality of virtual machines by the DPU, an instruction for creating a plurality of virtual machine managers to a host to which the DPU belongs, wherein one virtual machine manager is configured to manage one virtual machine; creating a plurality of virtual machine managers by the host, and concurrently starting a plurality of virtual machines by the virtual machine managers using a resource on the host; and offloading, in response to a completion of concurrently starting the plurality of virtual machines, the virtual machine managers from the host to the DPU, and managing the plurality of started virtual machines by the virtual machine managers using a resource on the DPU.