Dynamic Host-DPU Service Migration via Load Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CPU/DPU service provisioning systems face sub-optimal performance due to static mapping of services, which can lead to inefficient utilization of resources, particularly given the limited resources available to Data Processing Units (DPUs) compared to Central Processing Units (CPUs).
Innovation Solution
An Information Handling System (IHS) that includes a DPU processing system and a DPU memory system, which determines the operating loads of both the host subsystem and the IHS, and dynamically configures the host subsystem to provide services, allowing for proxying of service requests between client systems and the host subsystem, thereby optimizing resource allocation based on changing loads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a static/fixed mapping of services to CPU and DPU is provided, then service provisioning is simplified, but resource utilization becomes sub-optimal due to limited DPU resources
Solution Approach 1:
The patent implements dynamic service mapping that adapts to changing system conditions. The DPU monitors its own resource availability and the host CPU's workload state, then dynamically adjusts service placement decisions. This allows the system to transition from static to dynamic mapping, resolving the contradiction between provisioning simplicity and resource utilization efficiency.
Solution Approach 2:
The system incorporates feedback mechanisms where the DPU continuously reports its resource availability status and the host CPU reports its workload state. Based on this feedback, the service mapping is automatically adjusted to optimize resource utilization while maintaining operational simplicity through automated decision-making.
2Power
If services are offloaded to DPU, then CPU workload is reduced, but DPU resource constraints limit provisioning capability
Solution Approach 1:
The system dynamically adjusts the division of services between CPU and DPU based on real-time resource availability. When DPU resources are abundant, more services can be offloaded; when resources are constrained, the system adapts by adjusting the mapping dynamically, thus maintaining provisioning flexibility while respecting DPU constraints.
Solution Approach 2:
The system changes operational parameters such as service mapping configurations based on monitored resource states. By adjusting mapping parameters dynamically according to DPU and CPU resource conditions, the system optimizes the balance between CPU workload reduction and service provisioning capability.
3Device complexity
If DPU resources are limited, then hardware complexity is reduced, but service provisioning performance deteriorates
Solution Approach 1:
The DPU performs self-monitoring of its resource availability and automatically participates in service mapping decisions. This self-service capability allows the system to compensate for hardware limitations through intelligent resource management, maintaining service provisioning performance despite reduced hardware complexity.
Solution Approach 2:
The system uses dynamic service mapping to compensate for DPU hardware limitations. By adaptively adjusting which services run on the DPU versus the host CPU based on real-time conditions, the system optimizes performance despite the DPU's limited resource capacity.
Data Source
AI summary
A computing system includes a DPU subsystem coupled to a host subsystem. The DPU subsystem receives an instruction to provide a service, and determines a host operating load of the host subsystem and a DPU operating load of the DPU subsystem during subsequent time periods. Based on the host operating load and/or the DPU operating load during a first time period, the DPU subsystem configures the DPU subsystem to provide the service, and performs service requests for the service from client system(s). Based on the host operating load and/or the DPU operating load during a second time period that is subsequent to the first time period, the DPU subsystem configures the host subsystem to provide the service in order to migrate the service from the DPU subsystem to the host subsystem, and proxies service requests for the service between the client system(s) and the host subsystem providing the service.