一种基于DPU的主机侧延迟监测方法和装置

By implementing host-side latency monitoring through DPU and utilizing flow classification and explicit congestion notification strategies, this technology solves the problems of high host CPU overhead, low monitoring coverage, and poor robustness in existing technologies. It achieves efficient, low-overhead latency monitoring and anomaly management, making it suitable for modern data centers.

CN122420174APending Publication Date: 2026-07-17ZHEJIANG UNIV

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHEJIANG UNIV
Filing Date
2026-06-17
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing host-side latency monitoring solutions suffer from high host CPU overhead, low monitoring coverage, poor robustness in abnormal scenarios, high deployment costs, and poor compatibility, failing to meet the needs of modern data centers for low latency, high coverage, and real-time monitoring.

Method used

Leveraging the hardware characteristics of the DPU, TCP flows are classified through a flow classification strategy table. Combined with a token bucket sampling mechanism and a hardware timestamp module, full and selective tracking are achieved. A segmented explicit congestion notification strategy is used for real-time monitoring and anomaly feedback, completely offloading the host CPU load and achieving high coverage and robust monitoring.

Benefits of technology

It achieves host-side latency monitoring with zero host CPU overhead, high coverage, and high robustness. It is suitable for modern data centers, has strong compatibility, and is applicable to scenarios such as cloud computing, high-performance computing, and distributed storage. It has high practicality and promotional value.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122420174A_ABST
    Figure CN122420174A_ABST
Patent Text Reader

Abstract

本发明公开了一种基于DPU的主机侧延迟监测方法和装置,该方法包括将主机侧延迟监测逻辑完全卸载至数据处理单元DPU,利用DPU的可编程多核CPU、高速网络接口及加速引擎优势,实现监测与主机系统的硬件级隔离;通过对LS流和BE流分别采用全量覆盖监测和选择性抽样监测,在保证延迟敏感型业务监测精度的同时实现DPU资源均衡;建立延迟异常检测与拥塞反馈模型,通过对LS流延迟进行直方图统计,实时计算延迟均值、方差及99分位值,识别SLA违规,基于全局SLA压力得分采用分段式ECN标记策略,向BE流发送拥塞通知以缓解主机侧异常影响。本发明能够实现零主机CPU开销、高覆盖率、高准确性的主机侧延迟监测。
Need to check novelty before this filing date? Find Prior Art