Traffic demand estimation for hybrid switching systems

Abstract

Hybrid switching is considered to be a promising practical switching paradigm with good performance, high scalability and energy efficiency. One of the biggest challenges in designing such a switching system is mapping the incoming traffic appropriately onto the two types of switching resources with low complexity of hardware design. Existing work either utilizes simple traffic mapping heuristics, or makes idealistic assumptions on the incoming traffic patterns, and hence the problem of traffic demand estimation in hybrid switching systems is largely under-explored. In this paper, we evaluate the effectiveness of flow rate (FR) based and buffer occupancy (BO) based traffic estimation/mapping strategies. We then propose ISO-estimation, a novel traffic demand estimation method which is based on the two existing counters of typical switching systems with buffers, i.e. resource usage and queue length. The proposed method considers both traffic demand form new arriving packets and buffered packets. Simulation comparison on the performance of the proposed method against existing FR/BO methods and their combinations shows that the proposed method gives a better estimation result on traffic demand than the others, in terms of circuit switching packet percentage (CPP) and the number of circuit setup failures.