Lau Cheuk Lung’s research while affiliated with Universidade Federal de Santa Catarina and other places

In computing systems, some applications require high availability. The creation of copies improves availability, but keeping the copies synchronized requires the replication of the application state. Raft is a consensus algorithm that emerged with an easy understanding logic and a consequently well accepted solution. At infrastructure level, containers offer an alternative for replacing traditional virtual machines in cloud providers. This paper (This project was supported by CNPq proc. 401364/2014-3) evaluates the execution of Raft in physical machines and in Kubernetes, a container management system developed by Google and other companies. Results show similar performance for Raft in both environments.

Computer virtualization brought fast resource provisioning to data centers and the deployment of pay-per-use cost models. The system virtualization provided by containers like Docker has improved this flexibility of resource provisioning. Applications that require more restrictive agreement and ordering guarantees can also benefit from operating inside containers. This paper proposes the integration of coordination services in a container management system called Kubernetes (k8s), seeking to restrict the containers’ size and to offer automatic state replication. A protocol that uses shared memory available in Kubernetes was developed, and an evaluation was conducted to show the viability of the proposal.

State machine replication is an approach to create fault tolerant dis-tributed systems. The system remains corret while tolerating faults from someof its replicas. These replicas must execute the same sequence of requests, aproblem that is solved by total order algorithsm like Raft. some services mayneed that requests have different priority levels, such that some be executed be-fore others. In this work, we propose an algorithm, PRaft, that introduces theidea of priority based state machine replication to Raft, modifying it to deal withrequest priority.

State machine replication is a form of active replication commonly used to create fault-tolerant distributed services. In a nutshell, the approach consists in ensuring that a set of replicas receive and execute the same sequence of deterministic requests, returning the same results. This approach handles all requests evenly, but for some services it is important to consider that some requests have priority over others, i.e., that whenever two or more requests are ready to be executed, the one with higher priority is executed first. Paxos is perhaps the best known protocol to order requests in asynchronous environments, but Paxos has no notion of priority. In this paper we introduce the notion of priority-based state machine replication and modify Paxos to take request priorities into account. The proposed algorithm, PRaxos, works in three steps and satisfies Paxos' safety properties in asynchronous systems, while enforcing priorities when the system behaves synchronously.

Clouds are a suitable place to store data with scalability and financial flexibility. However, it is difficult to ensure the reliability of the data stored in a cloud. Byzantine fault tolerance can improve reliability, but at a high cost. This paper presents a technique that anticipates requests in order to reduce that cost. We show that this technique improves the performance in comparison with related works and maintains the desired data reliability.

A necessidade de compartilhar e manipular dados sensíveis é um desafio para grande parte dos provedores de conteúdo que utilizam a nuvem para armazenamento. No entanto, desenvolver aplicações que garamtam sigilo em nuvem é uma tarefa complexa e requer integração de múltiplos aspectos de segurança e interoperabilidade. Para contornar esses desafios, esse trabalho tem como principal objetivo propor uma arquitetura de middleware para garantir o compartilhamento seguro de documentos sigilosos em nuvem utilizando provedores de nuvens públicas para o armazenamento dos dados e módulos de segurança criptográficos para o gerenciamento de chaves criptográficas. Este trabalho tem como principais características: o uso da criptografia baseada em identidade, uso de nuvens híbridas, gerenciamento de chaves criptográficas simplificado, garantia de segurança ponto a ponto e utilização de módulos de segurança criptográficos.

Replication is often considered a cost-effective solution for building dependable systems with off-the-shelf hardware. Replication software is usually designed to tolerate crash faults, but Byzantine (or arbitrary) faults such as software bugs are well-known to affect transactional database management systems (DBMSs) as many other classes of software. Despite the maturity of replication technology, Byzantine fault-tolerant replication of databases remains a challenging problem. The paper presents MITRA, a middleware for replicating DBMSs and making them tolerant to Byzantine faults. MITRA is designed to offer transparent replication of off-theshelf DBMSs with replicas from different vendors.

Recently, researchers have shown an increased interest in concurrency control using distributed Software Transactional Memory (STM). However, there has been little discussion about certain types of fault tolerance, such as Byzantine Fault Tolerance (BFT), for kind of systems. The focus of this paper is on tolerating byzantine faults on optimistic processing of interactive and declared transactions using STM. The result is an algorithm named Mesobi. The processing of a transaction runs with an optimistic approach, benefiting from the high probability of messages being delivered in order when using Reliable Multicast on a local network (LAN). The protocol performs better when messages are delivered ordered. In case of a malicious replica or out-of-order messages, the Byzantine protocol is initiated.

A major challenge of computer systems is making these more robust, reliable and secure. In recent years, it has been found that the state machine replication is one of the most common techniques for designing systems that need high reliability. This paper introduces a new architecture for replication of Byzantine fault tolerant service, based on the tuple space model. The proposed model requires only 2f+1 replicas for one service, and is generic enough to accommodate a different set of services at the same execution infrastructure. Use of tuple space supports communication and coordination, as well as a storage medium for sharing of data between service replicas. In addition, the paper also presents an experimental evaluation of some protocols for replication of Byzantine fault tolerant systems for practical purposes.