Frequent elements on query defined ranges

Abstract

Heavy hitters and flow frequency measurements are fundamental in many networking applications. Further, some functionalities require insights about recent traffic, which is often captured by a sliding window. While existing works assume a single fixed size window, often multiple sizes are of interest and maintaining several instances is wasteful in terms of space and time complexity. Moreover, drill down analytics can also benefit from querying varying intervals. We propose efficient algorithms that support heavy hitters and frequency queries over an interval given at query time. Our methods asymptotically improve the naive approach of running multiple instances of fixed window size solutions. We evaluate our algorithm on a real Internet packet trace, showing that our methods are significantly more efficient than the naive approach and only pay a slight overhead compared to a (single) state-of-the-art fixed window algorithm.