Many applications in fields like sociology, biology and urban computing, need to cope with an explicit use of a spatial environment, or territory. Such applications, referred to as space-aware applications (SAAs), are based on a set of entities that live and operate in a territory. Parallel execution of space-aware applications is needed to improve the performance when the demand of computational resources increases. Despite the great interest towards SAAs, there is a lack of models and theoretical results for assessing and predicting their execution performance. This paper presents a novel framework, based on Stochastic Time Petri nets, which is able to capture the execution dynamics of parallel SAAs, and model the aspects related to computation, synchronization and communication. The framework has been validated by comparing the predicted performance results for a testbed application, i.e., the ant clustering and sorting algorithm, to those experienced on a real execution platform. An extensive set of experiments have been performed to analyze the impact on the performance of some important parameters, among which the number of parallel nodes and the ratio between computation and communication load.
Parallelization of space-aware applications: Modeling and performance analysis
Cicirelli Franco;Forestiero Agostino;Giordano Andrea;Mastroianni Carlo
2018
Abstract
Many applications in fields like sociology, biology and urban computing, need to cope with an explicit use of a spatial environment, or territory. Such applications, referred to as space-aware applications (SAAs), are based on a set of entities that live and operate in a territory. Parallel execution of space-aware applications is needed to improve the performance when the demand of computational resources increases. Despite the great interest towards SAAs, there is a lack of models and theoretical results for assessing and predicting their execution performance. This paper presents a novel framework, based on Stochastic Time Petri nets, which is able to capture the execution dynamics of parallel SAAs, and model the aspects related to computation, synchronization and communication. The framework has been validated by comparing the predicted performance results for a testbed application, i.e., the ant clustering and sorting algorithm, to those experienced on a real execution platform. An extensive set of experiments have been performed to analyze the impact on the performance of some important parameters, among which the number of parallel nodes and the ratio between computation and communication load.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.