Traffic matrices play a pivotal role in the management of an ISP's network such as various levels of traffic engineering and capacity planning. However, it is unclear how the interaction between the internal traffic routing policies chosen by the ISPs and large-scale content providers, and the ongoing trend of "cloud-computing" affect the traffic matrices. In this paper, we use network data collected by a Tier-1 ISP to understand the characteristics of a PoP-level traffic matrix. We also shed light on the role of "routing matrix" in shaping the characteristics of the traffic matrix. Two of the most important observations in this study are: a) multi-exit prefixes and use of early-exit routing are the major reasons why PoP level traffic for the large ISPs do not follow the gravity model(i.e. proportional distribution based on size) as assumed by previous works, and b) routing plays a fundamental role in shaping the traffic matrix.