Network function virtualization (NFV) is one of the key technology enablers for actualizing 5G networks. With NFV, virtual network functions (VNFs) are linked together as a service function chain (SFC), which provides network functionality for the customer on demand. However, how to efficiently find a suitable placement for VNFs regarding the given objectives is an extremely difficult issue. The existing approaches assume that the SFC has a simple and asymmetrical pattern that is unsuitable to modeling a real system. We address this limitation by studying a VNF placement optimization problem with symmetrical SFCs that can support both symmetric and asymmetric traffic flows. This NP-hard problem is formulated as a mixed-integer linear programming (MILP) model. An iterative greedy-based heuristic is proposed to overcome the complexity of the MILP model. Extensive simulation results show that the proposed heuristic can obtain a near-optimal solution compared to MILP for a small-scale network, and at the same time, is superior to a traditional heuristic for a large-scale network.