Du et al. have done similar work [12]. They propose a
contention matrix to characterize the flow congestion in a
global view. However, their limitations are: the contention
matrix uses only the traffic splitting proportion to represent
the traffic contention, which may result in inaccurate traffic
congestion; how to utilize the contention matrix to predict
congestion has not been well addressed, which may degrade
the importance of the contention matrix based on congestion
representation.