Dynamic Bare Accelerator Configuration for Server Resource Utilization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing accelerator configurations in servers lead to low resource utilization as each service type requires a dedicated accelerator, resulting in idle accelerators when not all services are actively used.

Innovation Solution

A method and apparatus that dynamically configure accelerators by determining the target service type and loading the corresponding service logic function onto a bare accelerator only when needed, allowing a single accelerator to support multiple service types and optimizing hardware usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If an accelerator is configured for each service type in advance, then the accelerator can provide acceleration service for the corresponding service type, but the accelerator resource utilization is low because some accelerators remain idle when not all services are actively used

Engineering Contradiction:
Improveservice acceleration capabilityVSAvoidaccelerator resource utilization
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent implements dynamic accelerator configuration where accelerators are not statically assigned to service types but are dynamically allocated based on real-time service demands. The system monitors service requirements and configures accelerators on-demand, transforming the static accelerator deployment model into a dynamic one that adapts to changing workload patterns, thereby improving resource utilization while maintaining service acceleration capabilities

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent makes accelerators universal by enabling them to serve multiple service types through dynamic reconfiguration. Instead of dedicating specific accelerators to specific service types, the system designs accelerators that can be programmatically configured to handle different service types as needed, allowing a single accelerator to perform multiple functions and serve various service types, thus eliminating idle resources

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If multiple accelerators are disposed in a server to support various service types, then all service types can receive acceleration services, but the device complexity and hardware resource requirements increase

Engineering Contradiction:
Improveservice type coverageVSAvoidnumber of accelerators
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent reduces device complexity by implementing universal accelerators that can be dynamically reconfigured to support multiple service types. Instead of deploying multiple specialized accelerators for different service types, the system uses a smaller number of versatile accelerators that can be programmatically adapted to handle various service types, thereby reducing the total number of hardware components while maintaining comprehensive service type coverage

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent employs dynamic configuration mechanisms that allow accelerators to change their functional characteristics based on service requirements. This dynamic adaptability enables a smaller number of accelerators to cover more service types by reconfiguring their operational modes and logic, eliminating the need for a large fixed inventory of specialized accelerators and thereby reducing device complexity

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3385835B1Method and apparatus for configuring accelerator
Publication Date: 2023.04.05 HUAWEI TECH CO LTD
  • EP3385835B1 patent drawingFigure 1~2
  • EP3385835B1 patent drawingFigure 3~4
  • EP3385835B1 patent drawingFigure 5

AI summary

The present invention relates to the field of computer technologies and discloses a method and an apparatus for configuring an accelerator, to increase accelerator resource utilization. The method is applied to a server including at least one bare accelerator. The bare accelerator is an accelerator that is generated after a basic logic function is loaded for accelerator hardware, and the basic logic function includes a communications interface function and a loading function. The method includes: determining, by the server, a target service type and a target bare accelerator; determining, by the server, a service logic function corresponding to the target service type; and loading, by the server, the service logic function corresponding to the target service type for the target bare accelerator, to generate a target service accelerator, where the target service accelerator is capable of providing an acceleration service for a service of the target service type.