Service performance tuning method and apparatus

By dynamically adjusting the memory access priority of high-traffic services, the problem of increased latency in autonomous driving SoCs was solved, achieving performance assurance for services with high latency requirements and efficiency improvement for AI services.

WO2026143339A1PCT designated stage Publication Date: 2026-07-09YINWANG INTELLIGENT TECHNOLOGIES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
YINWANG INTELLIGENT TECHNOLOGIES CO LTD
Filing Date
2024-12-30
Publication Date
2026-07-09

AI Technical Summary

Technical Problem

In autonomous driving SoCs, insufficient memory access priority adjustment for high-traffic services leads to increased latency, affecting the performance of services with high latency requirements. Furthermore, as AI services become more intelligent, the impact of static limiting parameters on AI service performance is gradually increasing.

Method used

The memory access priority of high-traffic services is dynamically adjusted by the service performance adjustment device. Based on the predicted bandwidth value, the memory access priority of high-traffic services is adjusted in advance according to the bandwidth occupation of high-traffic services, so as to ensure that services with high latency requirements occupy memory bandwidth and release resources when demand is low. Frequency domain analysis and event/fixed frequency triggering mechanism are used for precise adjustment.

Benefits of technology

It effectively reduces the bandwidth contention of high-traffic services on services with high latency requirements, improves the processing efficiency of AI services, and reduces the analysis and adjustment overhead caused by frequent adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024143813_09072026_PF_FP_ABST
    Figure CN2024143813_09072026_PF_FP_ABST
Patent Text Reader

Abstract

The present application discloses a service performance tuning method and apparatus. The method comprises: a service performance tuning apparatus determining at least one predicted bandwidth value of a first service within a first time period; and at a first time, adjusting a memory access priority of a second service according to the at least one predicted bandwidth value, wherein a latency requirement of the first service is higher than a latency requirement of the second service, and the first time precedes the first time period. After a bandwidth occupancy of a first service within a first time period is predicted online, a memory access priority of a second service (for example, a high-traffic service) can be adjusted at a first time according to the bandwidth occupancy of the first service within the first time period, allowing for the memory access priority of the second service to be adjusted in advance. This helps ensure that the first service can effectively occupy bandwidth when bandwidth occupancy is necessary, thereby satisfying bandwidth occupancy requirements for the first service, and enabling the memory access priority of the second service to be adjusted dynamically, so as to achieve effective tuning results.
Need to check novelty before this filing date? Find Prior Art