Distributed Data-Flow Service Framework Scheduling Feature Services
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solutions fail to integrate diverse AI/ML algorithms from different suppliers into a common platform for automotive systems, lacking flexibility in processing pipeline formation and resource optimization across distributed systems, including virtualized environments.
Innovation Solution
The Distributed Data-Flow Service (DDFS) framework schedules feature services within a SoC with standard and accelerator cores, using a generic wrapper for algorithm compatibility, flexible scheduling policies, and independent execution on standard or accelerator cores, enabling efficient deployment of classical and machine learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple AI/ML algorithms from different suppliers are integrated into a common platform, then algorithm compatibility and system versatility improve, but integration complexity and device complexity increase
Solution Approach 1:
The patent implements a universal interface layer that standardizes communication between diverse AI/ML algorithms and the host system. This interface layer provides common data structures, parameter formats, and control mechanisms that enable multiple algorithms from different suppliers to be integrated into a single platform without increasing integration complexity, thereby achieving universality across heterogeneous algorithm sources
Solution Approach 2:
The patent introduces an intermediary algorithm manager that acts as a mediator between the host system and multiple AI/ML algorithms. This manager handles algorithm registration, configuration, resource allocation, and coordination, shielding the host system from the complexity of integrating multiple proprietary algorithms while maintaining full versatility in supporting diverse algorithm implementations
2Productivity
If processing pipelines are formed in distributed systems with remote machines, then system scalability and processing capacity improve, but communication overhead and system complexity increase
Solution Approach 1:
The patent segments the AI/ML processing system into independent modular units that can be distributed across multiple machines. Each processing node operates autonomously with well-defined interfaces, allowing the system to scale by adding more nodes without proportionally increasing overall system complexity. The segmentation enables parallel processing while maintaining manageable system architecture
Solution Approach 2:
The patent employs an intermediary communication layer that standardizes data exchange between distributed processing nodes. This intermediary layer provides unified protocols for data transmission, synchronization, and coordination, reducing communication overhead and abstracting the complexity of distributed system management from individual processing nodes
3Productivity
If scheduling policies are configured for each feature service, then processing performance optimization improves, but configuration complexity and time consumption increase
Solution Approach 1:
The patent implements self-service scheduling where each feature service automatically configures and adjusts its own scheduling parameters based on pre-defined performance profiles and real-time system conditions. The services autonomously optimize their execution timing, resource allocation, and priority levels without requiring manual configuration, thereby achieving processing performance optimization while minimizing configuration time and complexity
Solution Approach 2:
The patent utilizes parameter-based scheduling configurations that allow feature services to dynamically adjust their execution parameters such as frame rate, processing priority, and resource allocation. These parameters can be modified at runtime based on system load and performance requirements, enabling flexible optimization without time-consuming reconfiguration
4Power
If AI/ML algorithms are executed on accelerator cores, then processing speed and energy efficiency improve, but hardware requirements and device complexity increase
Solution Approach 1:
The patent applies local quality by assigning specific AI/ML algorithms to appropriate accelerator cores based on their computational characteristics. Different types of accelerators (GPU, DSP, NPU) are selectively utilized for different algorithm types that best match their processing strengths, thereby optimizing processing speed and energy efficiency while avoiding the need to implement all types of accelerators in every system
Data Source
AI summary
The present disclosure generally relates to dataflow applications. In aspects, a system is disclosed for scheduling execution of feature services within a distributed data flow service (DDFS) framework. Further, the DDFS framework includes a main system-on-chip (SoC), at least one sensing service, and a plurality of feature services. Each of the plurality of feature services include a common pattern with an algorithm for processing the input data, a feature for encapsulating the algorithm into a generic wrapper rendering the algorithm compatible with other algorithms, a feature interface for encapsulating a feature output into a generic interface allowing generic communication with other feature services, and a configuration file including a scheduling policy to execute the feature services. For each of the plurality of feature services, processor(s) schedule the execution of a given feature service using the scheduling policy and execute a given feature service on the standard and/or accelerator cores.