Reatividade e qualidade de serviço em aplicações web

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

The huge success of the Internet has posed new challenges in terms of applications and user satisfaction. In order to meet the Quality of Service (QoS) expected by users, there are new requirements, such as performance and scalability, that must be fulfilled by the applications. Due to these requirements, QoS has become a relevant topic of interest. Several mechanisms to provide QoS have been proposed, but they usually do not consider aspects related to reactivity, e.g., how users react to variable server response time. This work addresses the use of reactivity to design novel QoS strategies. First, we demonstrate how the reactivity may be modeled and replicated using a methodology to correlate the measures of the response time and IAT (inter-arrival time) of user requests. Based on the model, we implement a new version of the httperf workload generator that is capable of reproducing the reaction of users. We evaluate the impact of a reactive workload based on the TPC-W benchmark to the performance of an actual Web server, and verified that reactive workloads are different from the non-reactive ones, affecting the server s load as well as its throughput and response times. In order to evaluate the impact of the reactivity and the new QoS strategies, we design a simulator of Internet applications named USAR-QoS. We prepare it to implement admission control and scheduling policies. Using the simulator, we evaluate the behavior of each user class as defined by the reactivity model. We then propose new admission control strategies capable of rejecting requests and sessions according to a reaction-based criteria. By simulating the new strategies we verify that they are effective to maintain the response time but may cause an increase in the request loss rates. We also present the PFIN (Patient-First Impatient-Next) and the IFPN (Impatient-First Patient-Next) scheduling strategies. We verify that they are effective for reducing the user request loss rates but may increase the response time. In order to optimize the benefits of each reactive approach we propose a hybrid multi-level approach that combines admission control and scheduling. The new QoS strategies are evaluated through simulation and compared to a scenario running the simple Best Effort FIFO scheduling approach and no admission control. The results demonstrate the effectiveness of the new policies and the hybrid multi-level approach presented to be the most effective mechanism to guarantee QoS considering the reactivity.

ASSUNTO(S)

telecomunicações trafego. administração teses. redes de computação confiabilidade teses. internet (redes de computação) avaliação teses. computação teses. redes de computação controle de qualidade teses.

ACESSO AO ARTIGO

Documentos Relacionados