Cloud Computing is gaining momentum every day as most companies are adopting cloud for their business. Cloud computing provides different kinds of service models: Infrastructure-as-a-Service, Platform-as-a-Service and Software-as-a-Service. Different organizations provide different solutions for each kind of cloud service model. Since the cloud services are used by most of the businesses, it needs to be very fast, accurate, easy to use, secure and fault tolerant. In this thesis, we improve the performance of a PaaS cloud called Bluemix, developed by IBM. Along with the performance, we improve the overall experience of the developers using a PaaS cloud for uploading their applications. Bluemix is used by developers to create, store, manage and run their applications. Developers create their application and push it to Bluemix to run the latest version of the application. We research on the algorithm behind the push mechanism of applications. This algorithm is know as the "file synchronization" algorithm. File Synchronization in PaaS cloud has three main components: Client, Server and Blobstore. These three components are connected to each other via network. Therefore, network latency and bandwidth are major factors deciding the performance of an approach for file synchronization. The goal of this thesis is to improve the overall user experience and performance of file synchronization in a PaaS cloud. To this end, we survey different solutions available for the file synchronization. One of the prominent examples is Rsync. We formally describe and evaluate the three suggested approaches for file synchronization in PaaS cloud in this thesis. We test the performance of the three approaches with different types of cache. We compare these approaches with the currently implemented approach. We measure the performance of an approach by measuring the total time of file synchronization, total bandwidth usage, total amount of storage used and the speed of synchronization. After evaluating all the approaches including the existing one, on the former metrics, we have successfully reduced the total time of synchronization, total storage and total bandwidth usage by a significant amount.