Algorithms for pipeline interval analysis of traffic

The problem of real-time queue analysis of production IP traffic is stated. The method of interval analysis is described in context of real-time constraints. Drawbacks of such method are stated and solutions proposed. First, we propose an algorithm to enhance interval analysis support for variable length analysis which is crucial for real IP-TV traffic. We consider different units to measure packet to customer in the queue mapping. We estimate the 1KB worth of packet data to be a viable fit to calculate accurate queue sizes. Then we propose pipeline extension for interval analisys as a sliding window. We introduce sliding window on timestamp scale and describe evolution models of left and right window edge separately. The algorithm proposed to map timestamps to number of customer in the queue. All proposed techniques then combined into algorithm of reverse queue calculation with parallel calculation of several service intervals simultaneously. We demonstrate and measure a performance of reference implementation of said algorithms in the lab under H.264 IP-TV traffic of various rates.