Batch request peak clipping processing method and system based on dynamic dormancy and partition polling

By employing a batch request peak-shaving method combining dynamic sleep and partitioned polling, this approach utilizes a reinforcement learning model to predict load trends, dynamically calculates the sleep time window, and performs content feature extraction and deduplication within the sleep window. This solves the problems of resource waste and system instability in high-concurrency scenarios, achieving efficient peak-shaving and resource optimization.

CN122086604APending Publication Date: 2026-05-26ZHEJIANG SHENGBAI INFORMATION TECH CO LTD
0 Cites 0 Cited by

Patent Information

Application Number
CN202610135394.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-01-30
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

In high-concurrency scenarios, traditional peak-shaving methods such as thread sleep, queue discarding, or fixed-frequency polling are difficult to cope with highly unpredictable peak pressure, leading to resource waste and system instability. In addition, a large number of semantically similar requests cause the server to perform far more operations than necessary.

Method used

A batch request peak shaping method based on dynamic sleep and partitioned polling is adopted. The load trend is predicted by the reinforcement learning model, the sleep time window is dynamically calculated, and content feature extraction, similarity clustering and content-level deduplication are performed within the sleep window to generate a deduplicated request set. The same type of request is processed uniformly to reduce the actual number of executions.

Benefits of technology

It significantly reduces processing rhythm fluctuations, improves system carrying capacity and stability, reduces resource waste, optimizes server resource utilization, adapts to business changes and load fluctuations, and achieves non-linear peak shaving effect.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a batch request peak clipping processing method and system based on dynamic dormancy and partition polling, and relates to the technical field of server-side task scheduling and load control. According to the method, firstly, batch requests are divided into a plurality of request partitions according to resource identifiers, and the queue length, the request arrival rate and the queue increase change rate of each partition are updated in real time; then, predicting the load trend of each partition in the next period through a reinforcement learning model, and calculating a dynamic sleep time window according to the load trend; and in the dormancy window, performing content feature extraction and similarity clustering on the to-be-processed requests, and realizing duplicate removal and consistency verification of the same-cluster requests in combination with content hash to form a final duplicate removal request set. And based on the processing feedback and the global peak clipping index, adaptive updating is carried out on the reinforcement learning strategy parameters, and cross-partition cyclic peak clipping control is realized. According to the method, the instantaneous load can be reduced in a high-concurrency scene, and the server resource utilization rate and the system stability are improved.
Need to check novelty before this filing date? Find Prior Art