Modern network functions (NFs) are able to scale out conveniently with the help of network function virtualization (NFV). When network traffic rises, NFV can provide more instances easily for migrating network flows to new instances to balance the traffic on the existing ones. However, the main challenge is to select which subset of existing flows to migrate when scale out, minimizing the negative impact of flow migration. In this paper, we propose the Co-Scaler, which is a heuristic algorithm to find out an optimized solution quickly. It employs two-step flow selection when a service function chain (SFC) with multiple NFs needs to scale out. We evaluated Co-Scaler by the real trace from data center network with several NFs. The results show that Co-Scaler can significantly decrease the number of packets affected by buffering, and minimize the latency introduced by flow migration.