Instantaneous Load Dependent Servers (iLDS) Model for Web Services

In the last few years, the paradigms used to create new business applications is deeply changed. As matter of fact, the service providers turned their old approach create from scratch whatever we need into a new one: look for and integrate. In this brand new scenario, the key role is played by the interoperability: indeed it is required that each time a new service is deployed and once the interfaces are defined, the client does not need any further information to make use of the service. The service oriented architecture (SOA) aims at providing a structure that guarantees the transparency and the interoperability between service providers and service requestors. Even if the SOA is widely diffused, its performances evaluation is still an open problem: due to transparency of the architecture, it is very hard for application/service developers to have quantitative performance evaluation at any development stages. In this context, we will give an answer to the following question: given a newly developed service, built in a closed development environment, how to predict its performance on a hosted service platform (for example in order to choice between two available platforms or to fine tune it in the contest of a large SOA application)? In this paper we propose a technique for semiautomatic building of performance models of Web service platforms, which we model as instantaneous load dependent servers (iLDS): servers whose service time depends on the workload at a given time. The approach we propose, adopts a black box measurement technique, this means that we can just deploy a service, but we have no access to the underlying server platform configuration both for software and hardware layers. The resulting model can be used by final user or service developers to predict the service behaviour respect to a known workload or to compare different web service platforms. The proposed approach was validated on a simple case study and the measurements put in evidence an inter- - esting result about the common behavior of instantaneous load dependent servers, which can be of general use for modeling this class of systems.