Efficiently Computing Latency Distributions by Combined Performance Evaluation Techniques

Abstract

Service-oriented systems are designed for interconnecting with other systems. The provided services face timing constraints, the so-called latencies. We present a high-level performance evaluation technique that can be used by a system designer to obtain distributions of these latencies. This technique is capable of capturing nondeterministic, probabilistic and real-time aspects in one go. Under the hood, the technique is equipped with two mechanisms: (i) selection of the right abstraction of the model (to prevent a state space explosion) by evaluating the performance of executing models of different complexities; and (ii) an efficient algorithm in which basic estimates, simulation, and (probabilistic) model checking are combined. We illustrate our approach with an case on image processing of interventional X-ray systems.

Evaluating load balancing policies for performance and energy-efficiency

“…To evaluate the performance and energy trade-off of many policies, we have implemented two load balancers with exactly the same specifications in iDSL [5][6][7][8] and in AnyLogic [1]. Alternatives for iDSL, which offers a high-level language, are PRISM [14], Modest [13] and UPPAAL [17].…”

“…E.g., when a server receives less incoming requests, it is more likely to go to sleep to save energy. Next, we present the specification of a load balancer with respect to power state behaviour in equations (4)(5)(6)(7)(8). Figure 1 visualizes the four power states as well as the roles the equations play with them, i.e., equation (4-6) are concerned with power three states, and equations (7) and (8) with one power state.…”

“…iDSL [5][6][7][8] is a formal language and solution chain to evaluate the performance of service-oriented systems. iDSL has been developed made using Eclipse for DSLs 5 and is, therefore, an Eclipse plugin with an extensive IDE.…”

Designing Cyber-Physical Systems with aDSL: a Domain-Specific Language and Tool Support

“…For another instance, a DSL is used to model serviceoriented systems [11]- [13]. After this, the model is automatically transformed into timed automata (TA) to compute absolute latency bounds [11], [14] and probabilistic timed automata (PTA) to compute latency distributions [15], [16] Freek van den Berg was supported by a 4TU. NIRICT for each service.…”