Prilagodljivo dinamičko raspoređivanje skupnih poslova na grozdu računala

doctoral thesis

• Title (english): Adaptive dynamic batch job scheduling in computer cluster
• Author: Igor Grudenić
• Mentor: Nikola Bogunović (mentor)
• Committee member: Domagoj Jakobović (predsjednik povjerenstva)
• Committee member: Nikola Bogunović (član povjerenstva)
• Committee member: Dragan Gamberger (član povjerenstva)
• Committee member: Leo Budin (član povjerenstva)
• Committee member: Siniša Srbljić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Electronics, Microelectronics, Computer and Intelligent Systems); Zagreb
• Defense date and country: 2010-12-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Universal decimal classification (UDC): 004 - Computer science and technology. Computing. Data processing

Abstract

U disertaciji su razmatrani različiti aspekti vezani uz raspoređivanje poslova na grozdovima računala. Istraživanje postupaka raspoređivanja poslova oslanja se na simulacije paralelnih računalnih sustava te je u sklopu rada oblikovan simulator koji omogućava aktivno sudjelovanje poslova u simuliranom sustavu. Provedena je analiza postojećih postupaka raspoređivanja poslova i izgrađen je učinkovit izvorni algoritam zasnovan na dinamičkom programiranju i postupku unazadnog popunjavanja praznina. Poslovi koji dolaze na računalne grozdove okarakterizirani su lošim procjenama trajanja, pri čemu je nezanemariv dio tih poslova neispravan. U svrhu poboljšanja učinkovitosti raspoređivanja analizirane su mogućnosti predviđanja neispravnih poslova i predviđanja trajanja poslova statističkim metodama za dubinsku analizu podataka. Dobiveni rezultati predviđanja iskorišteni su u modificiranim postupcima raspoređivanja te su izmjerena poboljšanja učinkovitosti tih postupaka.

Abstract (english)

This thesis addresses different aspects of computer cluster scheduling. Discrete event simulator that enables representation of jobs as active simulation entities is designed to enable research of scheduling algorithms. Theoretical and experimental analysis of the existing scheduling algorithms is made, which is accompanied with a designed of the new and efficient scheduling algorithm based on dynamic programming and backfilling. Jobs in computer clusters are characterized by inaccurate runtime estimates and ineligible amount of jobs fails while being executed. In order to overcome this, methods for runtime and failure prediction are designed and their efficiency is measured. These predictions are used to improve efficiency of cluster schedulers that are modified to accommodate them.

• Keywords: grozd računala; postupci raspoređivanja; trajanje poslova; neispravni poslovi; statističke metode za dubinsku analizu podataka
• Keywords (english): computer cluster; scheduling algorithms; runtime prediction; failure prediction; data mining
• Language: croatian
• URN:NBN: urn:nbn:hr:168:416173
• Project: Number: 036-0362980-1921; Title: Računalne okoline za sveprisutne raspodijeljene sustave; Leader: Siniša Srbljić; Jurisdiction: Croatia; Funder: MZOS; Funding stream: ZP
• Study programme: Title: Computer Science; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti (Doktor znanosti)
• Catalog URL: http://lib.fer.hr/cgi-bin/koha/opac-detail.pl?biblionumber=34353&query_desc=au%2Cwrdl%3A%20grudeni%C4%87
• Type of resource: Text
• Extent: 156 str ; 30 cm
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-05-07 08:04:54