I would like to thank my advisor and friend, Professor Alysson for all the help and encouragement he provided before and during the Master’s program. I certainly would not get this far without his advices and knowledge. I would like also to thank his wife and my friend Cassia for the support while we were in Portugal. I thank also my wife Fernanda for being there for me during this whole journey. She did not think twice when I asked her to move with me to Portugal for two years to work on the TClouds project. She also listened to me during all good and bad times I had during these two years. My mother in law Neusa, thank you for being such a great mother for Fernanda and also a friend to myself. My colleagues from FCUL helped a lot during this time, with answers, ideas and even digging with the code for problems I had. Thank you all for that. A special thanks to Vini-cus Cogo, without whom I would not be able to deliver this dissertation. Another thanks to Pedro Gonçalves who helped me with all the invoices, expenses reports, contracts and paperwork I needed in these two years. His dedication to his job is certainly something that Lasige could not work without.

Abstract—Cloud computing infrastructures leverage fault-tolerant and geographically distributed services in order to meet the requirements of modern applications. Each service deals with a large number of clients that compete for the resources it offers. When the load increases, the service needs to scale. In this paper, we investigate a scalability solution which consists in partitioning the service state. We formulate specific conditions under which a service is partitionable. Then, we present a general algorithm to build a dependable and consistent partitioned service. To assess the practicability of our approach, we implement and evaluate the ZooFence coordination service. ZooFence orchestrates several instances of ZooKeeper and presents the exact same API and semantics to its clients. It automatically splits the coordination service state among ZooKeeper instances while being transparent to the application. By reducing the convoy effect on operations and leveraging the workload locality, our approach allows propos-ing a coordination service with a greater scalability than with a single ZooKeeper instance. The evaluation of ZooFence assesses this claim for two benchmarks, a synthetic service of concurrent queues and the BookKeeper distributed logging engine. I.