Identifying global hot items in distributed dynamic data streams

Identifying hot items have been found useful in many network monitoring applications to detect network anomalies. There are different variants of this problem and we work on a special case where the traffic of the hot items are distributed in multiple dynamic data streams. Most of the existing methods for identifying hot items do not work for our problem, except one which used Group Testing based sketching algorithm to identify the hot items. In this paper, we propose a method to solve this special variant of hot item identification using sketches based on group testing and random projections. We will show that our method can identify hot items with higher accuracy and runs faster than previous methods, by both theoretical proofs and experimental evaluations. We will show how to tune the parameters used in our method which is a practical issue when it is used in practice.