Accurate Network Flow Measurement with Deterministic Admission Policy

摘要

Network management tasks require real-time visibility of current network status to perform the appropriate operations. However, the resource limitation of network devices and the real-time requirements make it difficult to provide accurate network measurement feedbacks. To reduce the error and inefficiencies caused by random operations in existing algorithms, we propose an efficient measurement architecture with the Deterministic Admission Policy (DAP). DAP provides accurate large-flow detection and high network measurement precision by making full use of the information belong to large flows and small flows, and dynamically filtrating small flows as the network status evolves. To make the algorithm easy to implement on hardware, we propose d-Length DAP by replacing the global optimality with local optimality. Experimental results show that our algorithm can reduce the measurement error by 3 to 25 times compared to other algorithms.