Gerenciamento de capacidade de infra-estruturas multicamadas com restrições de energia e ataques de segurança

Abstract

Along the last years, an increasing number of enterprises have turned to computational capacity outsourcing as a financially attractive approach to host their online services. This dissertation considers the scenario where many different third-party transactional Web services are hosted in a shared infrastructure. The focus is on the capacity management of these infrastructures, which consist in finding the most cost-effective way to allocate the available resources, aiming at maximizing the provider's business objective, i.e., total profit.The emergence of new demands by the clients impose additional operational problems to the capacity management of theses shared infrastructures. Furthermore, current Web services demand multi-tier hosting platforms, e.g., presentation, application and database tiers. The task of capacity management of such a multi-tier infrastructure is further challenging as a result of the higher system complexity. Capacity management of hosting infrastructures have traditionally focused on performance objectives. However, quality of service provided to the hosted applications and, ultimately, the provider's profit, depend also on other aspects, such as security attacks and energy constraints.This dissertation extends a previous framework based on a single-tier infrastructure model to capture, more accurately, the main trade-offs of multi-tier infrastructures. The solution presented also captures, in an unified framework, the key performance and cost trade-offs that arise when operating under security attacks and energy constraints. The main contributions are: (1) a more accurate multi-tier performance model, that captures application-specific bottlenecks and the inherent parallelism of multi-tier platforms, (2) a flexible business model, with adaptive contracts for periods when security attacks or energy constraints are in effect, and (3) a new and much more complex optimization model, combining both performance and business models, responsible for calculating the capacity allocation decisions.The new approach is evaluated through simulation experiments with synthetic and realistic workloads in various scenarios. The results show that the new multi-tier solution is far more cost-effective, in terms of the provider's attained profit, than the previous approach, which it is built from, that employs a single-tier performance model. It also significantly outperforms static capacity allocation for heavy and heterogeneous workloads. The framework can minimize the impacts of security attacks and energy constraints through conscious capacity allocation decisions. Finally, adaptive contracts can be used to find a compromise between the provider's and its clients' interests.