Hierarchical Video Query Configuration via Pareto Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video analytics solutions are inefficient in resource allocation for processing live video streams, often resulting in low accuracy due to conservative resource demands and underutilization of available resources, particularly in hierarchical cluster environments where the uplink bandwidth between private and public cloud services is insufficient.
Innovation Solution
The proposed solution involves determining the most promising configurations of video query components by applying Pareto efficiency concepts to filter out inaccurate configurations with high resource demands, merging common components, and using a heuristic to search within the Pareto band for higher accuracy while efficiently utilizing resources, defined by a dominant utilization metric.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all video queries are run at the highest accuracy setting, then detection accuracy is maximized, but the private cluster does not have enough compute capacity and the uplink bandwidth is insufficient to push all streams to the cloud
Solution Approach 1:
The system dynamically adjusts query configurations and component placements based on real-time resource availability and accuracy requirements. Instead of static high-accuracy settings, the system adapts configurations to match current cluster capacity and bandwidth conditions, allowing optimal operation under varying resource constraints.
Solution Approach 2:
The system changes multiple parameters simultaneously including frame resolution, detector implementation choices, and component placement locations across the hierarchy. By adjusting these parameters based on resource availability, the system achieves near-optimal accuracy while staying within compute and bandwidth constraints.
2Productivity
If video query components are placed conservatively to reduce resource demands, then resource utilization is improved, but detection accuracy decreases
Solution Approach 1:
The system adds the placement dimension to the configuration space, allowing components to be distributed across multiple levels of the hierarchy (camera, private cluster, public cloud). This multi-dimensional placement strategy enables efficient resource utilization by placing computationally intensive components where resources are available while maintaining accuracy through strategic component distribution.
Solution Approach 2:
The system segments video query processing across the hierarchical cluster structure, dividing components between camera-based processing, private cluster processing, and public cloud processing. This segmentation allows each level to operate within its resource constraints while collectively achieving high accuracy through coordinated processing.
3Productivity
If the uplink bandwidth between private cluster and public cloud is increased to stream all camera feeds, then cloud-based analytics capacity is maximized, but the cost and complexity of infrastructure increases
Solution Approach 1:
The system extracts only the essential video data and analytics results that need cloud processing, rather than streaming all camera feeds. By filtering and selecting only necessary data for cloud transmission, the system maximizes cloud analytics capacity utilization while minimizing network bandwidth requirements and infrastructure complexity.
Solution Approach 2:
The system performs preliminary processing of video streams at the camera and private cluster levels before sending data to the cloud. By pre-processing videos to extract key features and filter relevant information, the system reduces the bandwidth required for cloud transmission while maintaining analytical capability.
4Productivity
If multiple queries share common components, then resource utilization is improved, but merging is restricted by requirements that components must have the same plan and be placed in the same cluster
Solution Approach 1:
The system creates universal component instances that can serve multiple queries with different configurations. A single detector component can be shared across multiple queries with different frame rates or resolution requirements, allowing resource consolidation while maintaining query-specific performance characteristics through configurable parameters.
Solution Approach 2:
The system introduces an intermediary layer that manages component sharing across queries with different plans and placements. This intermediary coordination mechanism allows components to be shared while respecting query-specific requirements, bridging the gap between resource efficiency and configuration flexibility.
Data Source
AI summary
A global manager communicates with various local managers to receive and process video queries. The video queries identify components that process live video streams, placement options for where the components of the video query may be executed, and various video query plans. The video query plans include options such as framerate and video quality. As the global manager processes the video queries, the global manager determines an initial set of video query configurations that identify a video query plan and placement option for each component of a given video query. Using the initial set of video query configurations, the global manager then determines an optimal set of video query configurations for the received set of video queries. The global manager communications instructions to the local managers to execute the components of the video queries using the video query plans and placement options from the optimal set of video query configurations.