BARBOSA, A. C. B.; http://lattes.cnpq.br/1574811347632895; BARBOSA, Ana Carolina Benjamim.
Resumo:
Service level management (SLM) of Web and Grid Services is an important issue that has to be solved in order to achieve large-scale deployment of services. The need to negotiate, measure and audit the quality of the services provided (in terms, for instance, of performance, cost or security) will increase directly with the proliferation of services. SLM comprises the following phases: (i) Service Level Agreement (SLA) negotiation and establishment; (ii) SLA auditing (iii) notification of SLA violations to the involved parties; and (iv) triggering of management actions in order to correct services to avoid new violations or to apply penalties to the assigning parties that did not comply with the SLA. Although some research efforts address SLM automation to Grid and Web Services, they do not detail the SLA auditing process. This dissertation focuses on the SLA auditing phase. We here present seven architectures that perform SLA auditing, as well as a qualitative and quantitative evaluation of such architectures. The qualitative evaluation compare the architectures based on aspects such as intrusiveness, trust, use of extra requests, possibility of preferential treatment, possibility of auditing of Consumer load, and possibility of auditing of encrypted messages. The quantitative evaluation focuses on the performance penalty that auditing introduces. In this evaluation, two factors that reduce service performance in the Client view due to auditing are described. One is the sample size of the extra requests sent by Inspector and obtained by the Provider in order to calculate the (Service Level Indicator) SLI values. The greater this sample size, the greater is the impact on service performance. Another factor is the Measurement Interference Error (MIE), that is, the error introduced in the measured value due to the observation method itself. These two factors are analyzed in the seven architectures and the MIEs are analyzed in more details by means of experiments performed with an implementation of some of these architectures.