In the context of increasingly prevalent data-intensive computing, cross-regional task collaboration has become the norm. A coflow typically represents a set of parallel data flows connecting different stages of a job across data centers, and its completion signifies that the job can proceed to the next stage. In multi-data center environments, ensuring coflow transmission performance while controlling cost and achieving link load balancing has become a pressing core issue. Although numerous coflow schedulers have been proposed, most focus only on a single performance objective. From the perspective of multi-objective optimization, this paper jointly optimizes routing paths and bandwidth allocation, and proposes an online optimization framework called Tongs. The framework can adaptively coordinate network resources in dynamic environments to balance performance, cost, and load. Tongs constructs an ideal single-coflow scheduling scenario and applies linear relaxation to obtain a near-optimal solution under non-competitive network conditions. Then, based on the weight factors derived from the single-coflow solution, the framework dynamically adjusts bandwidth allocation to enable online scheduling of multiple coflows. Large-scale simulation experiments further verify the framework’s remarkable capability in balancing performance, cost, and load, while significantly accelerating overall job completion.