Accurate measurement of network traffic is an essential part of current network management tasks. Traditional measurement methods based on counter and sketch respectively focus on the large flow detection and all flows queries, and the measurement accuracy is limited by the uncertainty in the large flow detection algorithm. Therefore, we design an asymmetric measurement architecture named DAP-Sketch. By using different data structures to measure large and small flows separately, DAP-Sketch provides the ability to query the approximate sizes of small flows while ensuring large flow detection. We propose a Deterministic Admission Policy (DAP) to dynamically distinguish the large and small flows, which effectively improves the accuracy of large flow detection and reduces the demand for storage resources. To improve the usability and universality of DAP, we put forward a d-Length DAP which applies local optimality instead of global optimality and makes our algorithm easy to implement. Furthermore, two optimization strategies with adaptive parameter adjustment are also designed in terms of the changes in memory space and traffic characteristics. Experimental results show that DAP-Sketch can outperform the best of the five typical measurement methods up to 59.3 times, decrease the requirements of network equipment resources, and achieve highly-precise and low-overhead network traffic measurement.

准确测量网络流量是当前网络管理任务的重要组成部分。传统的基于计数器和草图的测量方法分别专注于大流量检测和所有流量查询，并且大流量检测算法的不确定性限制了测量精度。因此，我们设计了一个名为DAP-Sketch的非对称测量架构。通过使用不同的数据结构分别测量大流量和小流量，DAP-Sketch可以查询小流量的近似大小，同时确保检测大流量。我们提出了一种确定性准入策略（DAP），以动态区分大流量和小流量，从而有效地提高了大流量检测的准确性，并减少了对存储资源的需求。d-长度DAP，它应用局部最优而不是全局最优，并使我们的算法易于实现。此外，还针对内存空间和流量特性的变化设计了两种具有自适应参数调整的优化策略。实验结果表明，DAP-Sketch可以胜过5种典型测量方法中的最佳方法，达到59.3倍，减少了对网络设备资源的需求，并实现了高精度和低开销的网络流量测量。