OptSCORE: Self-optimised efficiency, security and resilience for active replication

点击翻译按钮获取中文摘要

Reliable IT services are essential for a modern society. Networked systems have to solve a multitude of problems pertaining to availability, safety and security demands. Replication is one basic method to tackle these problems. Cloud computing, service-oriented architectures and multicore CPUs challenge established replication architectures due to their high parallelism and dynamicity.The goal of the OptSCORE project is to find new solutions for these challenges. To this end, we focus on State-Machine Replication (SMR) as the core architecture. SMR-based replicated services provide high availability coupled with strong consistency, and can tolerate not only crashes but even Byzantine faults.After the first project phase determined configuration parameters for deterministic scheduling and group communication systems, our next goal is to find configurations of these parameters for any given application, so that throughput and/or latency become optimal. The challenge lies in the complexity of possible parameter combinations and their interdependencies. We plan to use combinatorial optimisation methods as well as machine learning for finding optimal configurations. Additionally, we want to enable SMR systems to dynamically self-adapt to a replicated application's demands during runtime.Furthermore, we want to augment the inherent fault masking in SMR systems with intrusion-detection mechanisms to integrate error correction methods, especially in Byzantine fault-tolerant installations. A comprehensive security analysis will yield attacks both on SMR systems in general and on the aforementioned optimisation mechanisms. Researching suitable defensive mechanisms is a main goal of this research project.Several further optimisation strategies complement these efforts. In particular, we will try to research and solve the following problems: Do CRDTs allow to improve efficiency by employing partially as opposed to totally ordered multicasts, without losing strong consistency? How can a slow replica be prevented from slowing down the entire system? How can the necessary checkpointing mechanisms be made to work with concurrent processing of requests?The expected research results of this projects will significantly contribute to making highly-available and reliably services more efficient and to simplify and accelerate their configuration.