This paper addresses the issue of efficient and accurate petformance prediction of large-scale message-passing applications on high petforniance architectures using sinidation. Such simulators are often based on parallel discrete event simulation, Qpically using the conservative protocol to synchronize the simulation threads. The paper considers how a compiler cat1 be used to autoniatically extract information about the lookahead present in the application, and how this can be used to improve the performance of the null protocol used for synchronization. These techniques are implemented in the MPI-Sin1 siniulator and dHPF compiler, which had previously been extended to work together for optimizing the sinidation of local coniputational components of an application. The results show that the availability of lookahead iiforniation iniproves the runtinie of the siniulator by factors rarigitig from 9 % up to two orders of niagnitude, with 30.60% iniprovenietits being typical for the real-world codes. The experiments also show that these iniprovements are directly correlated with reductions in the number of ruill niessages required by the simulations. 1

Abstract not found

this paper, we propose and evaluate practical, automatic techniques that exploit compiler analysis to facilitate simulation of very large message-passing systems

apport de recherche

apport de recherche