These days we are witnessing a tremendous increase in the popularity of wireless devices, e.g. smartphones and tablets. These devices are typically connected to the Internet through cellular connections, such as LTE/4G. Because of the popularity of the wireless devices, a large portion of the traffic on the Internet goes through the cellular base stations. Caching the contents at the base stations brings the contents closer to the users, reduces the traffic on the Internet, and reduces the cost of providing the contents. In this paper, we study the problem of collaborative caching in cellular networks among a set of base stations. Motivated by the emergence of cloudlets, we consider unlimited cache space in our model, and our objective is to minimize the aggregated caching and download cost. We show that in the case of knowing the popularity of the contents, this optimization has a submodular property, and a greedy algorithm can achieve an approximation ratio of 2 for this optimization. We also provide an online algorithm that does not require any knowledge about the future requests and the content popularity. In order to evaluate our online algorithm, we compare its performance against the optimal solution through simulations.