Network function virtualization is a new network concept that moves network functions from dedicated hardware to soft-ware-defined applications running on standard high volume severs. In order to accomplish network services, traffic flows are usually processed by a list of network functions in sequence which is defined by service function chain. By incorporating network function vir-tualization in inter-data center (DC) network, we can use the network resources intelligently and deploy network services faster. However, orchestrating service function chains across multiple data centers will incur high deploy-ment cost, including the inter-data center bandwidth cost, virtual network function cost and the intra-data center bandwidth cost. In this paper, we orchestrate SFCs across multi-ple data centers, with a goal to minimize the overall cost. An integer linear programming (ILP) model is formulated and we provide a meta-heuristic algorithm named GBAO which contains three modules to solve it. We implemented our algorithm in Python and performed side-by-side comparison with prior algorithms. Simulation results show that our proposed algorithm reduces the overall cost by at least 21.4％ over the existing algorithms for accommodating the same service function chain requests.