Abstractions for Collective Adaptive Systems

Comparing perfomance abstractions for collective adaptive systems

“…In this paper, we apply such approach by adopting (i) a different modelling language, hinging on behavioural specifications and (ii) relying on queueing networks [20] for performance analysis. The modelling language used here has been advocated in [18] for specifying global behaviour of CAS. As shown in [18], this modelling language has a natural connection with queueing networks, therefore enabling performance analysis of CAS.…”

“…Using a simple, yet representative, robot scenario, we apply two different approaches to the performance analysis of CAS. Besides showing how to use behavioural specifications to analyse non-functional properties of CAS (emergent) behaviour, this exercise is instrumental for our contribution, which is a study of the relation between two complementary approaches to the performance analysis of CAS recently proposed respectively in [31] and in [18]. The quantitative analysis in [31] is based on generalised stochastic Petri nets, while the one in [18] relies on queuing networks.…”

“…The approach in [31] requires that the designer must directly come up with a performance model (rendered as generalised stochastic Petri nets). Instead, in the latter approach [18], the designer does not have to directly develop the queueing network for the quantitative analysis, because it is indeed 'compiled' from the behavioural specification of the CAS. In this sense, the former approach is a model-based methodology, while the latter is a language-based methodology.…”

“…RQ1 To what extent the approaches in [18] and in [31] support performance-aware design of CAS? RQ2 How do the features of the approaches in [18] and in [31] compare?…”

“…independent, collaborative and centralised) for this scenario. Section 3 provides the models based on the specification language in [18] for the three architectures. Section 4 shows the performance analysis based on the proposed models of Sect.…”