Cluster systems and simulation: from benchmarking to off‐line performance prediction

Abstract

SUMMARYThis paper describes a simulation-based technique for the performance prediction of message-passing applications on cluster systems by means of benchmark data. Given data measuring the performance of a target cluster in the form of standard benchmark results, along with the details of the chosen computing configuration, it is possible to build and to validate automatically a detailed simulation model. This makes it possible to predict off-line, i.e. without resorting to the real hardware, the performance of fully developed or even of skeletal code. An XML-based language (MetaPL) is adopted to describe the application behavior in the development stage. After a description of the approach and the illustration of the construction and validation of the simulation model, the paper presents a case study.

Autonomic Composite-service Architecture with MAWeS

“…These traces are amenable to direct simulation in HeSSE, thus making it possible to obtain fairly reliable predictions of program performance, even at the very first steps of code development [13]. The MAWeS framework relies on MetaPL descriptions, describing the software structure, and on HeSSE configuration files, describing the hardware/software execution environment, to run HeSSE simulations and to obtain performance data.…”

Planting parallel program simulation on the cloud

“…The application of simulation techniques to the performance prediction of distributed‐memory parallel programs dates back to the early nineties. In addition to the work by some of the authors of this paper, early efforts include the execution‐driven MPI‐Sim and the trace‐driven Dimemas network simulator. Among the most recent contributions in the literature, Phantom is a hybrid prediction framework that uses direct execution of computational blocks together with communication traces.…”

“… Most of our past results have been obtained by running a program simulator fed with actual or synthetic traces. Though decidedly faster than actual program executions, simulation of a complex code running on a high number of processors is a time‐consuming process. The construction of performance diagrams of a given code (most of the times, the evaluation of speedup for progressively higher number of processors until the knee of the curve is reached) requires many simulations.…”

Cloud-Based Concurrent Simulation at Work: Fast Performance Prediction of Parallel Programs

“…In addition to the work by some of the authors of this paper ( [6], [10]), early efforts include the execution-driven MPI-Sim [22] and the trace-driven Dimemas network simulator [23]. Among the more recent contributions in literature, Phantom [24] is a hybrid prediction framework that uses direct execution of computational blocks together with communication traces.…”

“…Our research group has shown that this is possible through the simulation of the program behavior [6], [7]. Most of our past results have been obtained by running a program simulator fed with actual or synthetic traces [8], [9], [10], [11]. Though decidedly faster than actual program executions, simulation of a complex code running on a high number of processors is a lengthy process.…”