In this Master’s Thesis the author proposes an innovative benchmark with the capabilities of reproducing a specific workload, set by the user, and computing some metrics that allow the analysis of the generated traffic. The tool aims to allow the modeling and the optimization of an arbitrary application by simulating its behavior in a given system and verifying its performance. After a brief introduction to the reasons that led to the development of such a tool, this work first describes the cutting edge technologies of benchmarks, parallel and distributed systems, secondly it provides the basic information that is necessary for the understanding of the methods and tools used in the dissertation. Then, it details the developed benchmark and its components, presenting its architecture, its working principles, its implementation and the produced analyses. Finally, it demonstrates the ability to reproduce the workload of an existing, recognizable and publicly accessible application. The experimental validation models a program implemented for the Apache Hadoop framework, using the MapReduce programming paradigm, after having analyzed its characteristics and extracted a possible taskgraph.
In questa Tesi di Laurea Magistrale l’autore propone un benchmark innovativo con la capacità di riprodurre un workload determinato, impostato dall’utente, e di calcolare alcune metriche che consentono l’analisi del traffico generato. Lo strumento sviluppato ha l’obiettivo di consentire la modellazione e l’ottimizzazione di un’applicazione arbitraria, simulandone il comportamento in un sistema e verificandone le prestazioni. Dopo una breve introduzione ai motivi che hanno spinto allo sviluppo di un simile strumento, il presente elaborato introduce dapprima il contesto nel quale l’applicativo si colloca, descrivendo lo stato dell’arte di benchmark, sistemi paralleli e sistemi distribuiti, per poi fornire le nozioni di base necessarie alla comprensione dei metodi e degli strumenti utilizzati. In seguito, si dettaglia il benchmark in ogni sua parte, presentando la sua architettura, il suo funzionamento, la sua implementazione e le analisi prodotte. Infine si dimostra la sua capacità di riprodurre il workload di un’applicazione esistente, nota e pubblicamente accessibile. La validazione sperimentale modella un programma implementato per il framework Apache Hadoop mediante il paradigma MapReduce dopo averne analizzato le caratteristiche e ricavato un possibile taskgraph.
Generazione di workload per sistemi paralleli
ROSA', ANDREA
2012/2013
Abstract
In this Master’s Thesis the author proposes an innovative benchmark with the capabilities of reproducing a specific workload, set by the user, and computing some metrics that allow the analysis of the generated traffic. The tool aims to allow the modeling and the optimization of an arbitrary application by simulating its behavior in a given system and verifying its performance. After a brief introduction to the reasons that led to the development of such a tool, this work first describes the cutting edge technologies of benchmarks, parallel and distributed systems, secondly it provides the basic information that is necessary for the understanding of the methods and tools used in the dissertation. Then, it details the developed benchmark and its components, presenting its architecture, its working principles, its implementation and the produced analyses. Finally, it demonstrates the ability to reproduce the workload of an existing, recognizable and publicly accessible application. The experimental validation models a program implemented for the Apache Hadoop framework, using the MapReduce programming paradigm, after having analyzed its characteristics and extracted a possible taskgraph.File | Dimensione | Formato | |
---|---|---|---|
2013_10_Rosà.pdf
Open Access dal 11/09/2014
Descrizione: Testo della tesi
Dimensione
1.37 MB
Formato
Adobe PDF
|
1.37 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/85164