Network Request Coalescing With Bloom Filters for Duplicate Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The rapid growth of programmatic advertising has overwhelmed network resources, causing network delays, poor performance, inefficient ad placement, and excessive computing costs due to the overwhelming volume of duplicate ad request streams, which existing caching systems are unable to handle efficiently within the required millisecond time frame.
Innovation Solution
A system utilizing a shared nothing architecture with independent coalescing machines applies consistent hashing to categorize and process request groups, identifying primary requests and holding duplicates until the primary is identified, operating at machine in-process speeds to eliminate duplicates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If caching systems are used to find and coalesce duplicate requests, then duplicate removal capability is improved, but processing speed deteriorates because caching systems are too slow to operate at machine speed
Solution Approach 1:
The patent replaces traditional caching systems with a Bloom filter data structure that operates entirely in memory without disk I/O or complex cache management. The Bloom filter provides probabilistic duplicate detection with O(1) time complexity, enabling machine-speed operation while maintaining effective duplicate removal capability.
Solution Approach 2:
The patent changes the operational parameters by using in-memory bit arrays and hash functions instead of traditional cache storage and retrieval mechanisms. This parameter change enables the system to process requests at machine speed (microseconds) rather than cache speed (milliseconds), resolving the speed-reliability contradiction.
2Stability of the object's composition
If centralized out-of-process database systems are used to handle request streams, then data consistency is improved, but processing speed deteriorates due to inter-machine communication latency
Solution Approach 1:
The patent segments the monolithic centralized database system into distributed in-memory coalescers that operate independently. Each coalescer maintains local state in RAM, eliminating inter-machine communication for duplicate detection. This segmentation provides both the speed of local processing and the consistency of coordinated duplicate removal across the distributed system.
Solution Approach 2:
The patent introduces Bloom filters as an intermediary data structure that enables fast probabilistic duplicate detection without requiring centralized database queries. The Bloom filter acts as a mediator between request streams and the coalescing logic, providing O(1) lookup speed while maintaining data consistency through coordinated updates across distributed instances.
3Measurement precision
If DSP servers evaluate enormous volumes of traffic to handle RTB ad exchanges, then ad placement accuracy is improved, but computing costs and scalability deteriorate
Solution Approach 1:
The patent extracts the duplicate request identification and removal function from the DSP server evaluation process. By pre-filtering duplicate requests using Bloom filters before they reach the DSP evaluation logic, the system reduces traffic volume by eliminating redundant processing, thereby lowering computing costs while maintaining ad placement accuracy for unique requests.
Solution Approach 2:
The patent performs preliminary duplicate detection and removal using Bloom filters before requests reach the expensive DSP evaluation stage. This preliminary action filters out redundant traffic upfront, allowing the DSP servers to focus computational resources only on evaluating unique, non-duplicate requests, thus improving scalability and reducing computing costs.
Data Source
AI summary
A system and method for coalescing network request streams assigns groups of requests to independent coalescing machines, each of which separately processes its assigned group according to processing rules. The use of independent coalescers allows the system to be scaled nearly infinitely with a shared nothing architecture and operates fully at machine in-process speeds.