Data analytics tends to be geo-distributed, which means the intermediate data may distribute into different data centers. It requires tenants to launch coflows across WAN links. However, there is a gap between tenants and the SD-WAN controller. This gap makes existing SD-WAN bandwidth allocation mechanisms drive to bad performance at tenant level. To fit this gap, this paper presents COC, a novel SD-WAN bandwidth allocation framework using hierarchical coflow ordering, to efficiently improve the coflow completion time of each tenant.