ARMHH: Accurate, Rapid and Memory-Efficient Heavy Hitter Detection with Sliding Window in the Software-Defined Network Context

Abstract

Network measurement is essential to many network applications. The trend of ever-increasing network traffic calls for contemporary network measurement approaches to have three key characteristics: accuracy, timeliness, and memory efficiency. The sketch algorithm offers a suitable trade-off between accuracy and memory consumption; however, its interval method is not as accurate and rapid as the sliding window method in Heavy Hitter detection. However, the sliding window method is believed to be highly memory-consuming. Accordingly, this paper proposes ARMHH: Accurate, Rapid, and Memory-efficient Heavy Hitter detection, which combines the sliding window and sketch algorithm approaches. Compared with the interval method, our experiments show that ARMHH improves the accuracy by 121.8x, 2.8x, and 3.5x on average for CM, CU, and Hashpipe, respectively, by consuming additional memory overhead for the sliding window. ARMHH also compresses the sliding window memory by at most 60%.