Recently, Mobile Edge Computing (MEC) is proposed to deploy with cellular base stations (BSs) to reduce the offloading delay and to provide computational resources and storage to nearby mobile users. However, the deployment of MEC with a BS introduces special challenges to mobility management. First, a MEC can only be accessed within the coverage area of its BS. Therefore, a user needs to perform both a radio handoff (HO) and a service migration when moving out of the cell coverage range. Second, as MECs have limited computational resource and storage, the target MEC may not have sufficient available resources to support new users. Therefore, a user with computation task may be forced to perform a radio HO to a different target BS in order to successfully migrate the service. These cause unnecessary HOs in cellular networks, especially when MECs are deployed with small cells, e.g., femtocells. These unnecessary HOs and service migrations incur a large signaling and migration cost. In this paper, we first propose a novel architecture, SharedMEC, to support user mobility. In addition, we propose a service HO decision algorithm and an analytical model to analyze the total cost which considers the total HO signaling cost, total migration signaling cost, and total migration cost. Simulation results show that our proposed architecture and service HO decision algorithm can significantly reduce the total cost. To the best of our knowledge, this is the first work that considers HO and migration issues together and analyzes the total cost in MEC systems.