Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Association Mining and Prediction of SystemPerformance Attributes in a large-scale ITinfrastructure
KTH, School of Computer Science and Communication (CSC).
2013 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

The master thesis seeks to establish relationships between

eleven system performance metrics in a large-scale IT infrastructure,

and to predict their future behavior. The rst

task was performed using multiple linear regression, with

one of the eleven metrics being taken as a dependant variable

measuring total system performance. Results of regressions

run over variations of the source data were evaluated,

and two of the metrics were concluded to be of consistently

high ability to explain the variability in the metric

measuring total system performance. The second task of

prediction was approached by attempting to replicate the

results from another research paper which presented a similar

problem and relevant results. A Kalman lter calibrated

by expectation maximization was used alongside vector autoregression

to evaluate the possibility of doing predictions.

The results were not found to be of obvious practical use,

with the exception of the vector autoregression procedure

which highlighted regularities in the metrics. When unifying

the sampling rate of the performance metrics, additional

descriptive statistics over the sampling intervals were

extracted when downsampling in order to retain potentially

useful information, which was found by manual inspection

in the case of one of the metrics. The inclusion of these

additional statistics was, however, not found to have a positive

impact in the regression analysis or in the prediction

attempts.

Abstract [sv]

Examensarbetet försöker fastställa samband mellan elva

variabler som mäter systemprestanda i en storskalig ITinfrastruktur,

och prediktera deras framtida beteende. Den

första uppgiften utfördes genom att använda multipel linjär

regression, där en av de elva variablerna var den beroende

variabel som mäter systemets viktigaste prestanda. Resultaten

av regressionskörningar över varianter av källdatan

utvärderades, och två av variablerna fastslogs ha konsekvent

hög förmåga att beskriva variationen i den beroende variabeln.

Den andra uppgiften rörande prediktion angreps genom

att försöka replikera resultaten från en vetenskaplig artikel

som presenterade ett snarlikt problem och för oss relevanta

resultat. Ett Kalmanlter kalibrerades genom förväntansmaximering

(expectation maximization) och användes tillsammans

med vektorautoregression för att utvärdera möjligheten

att göra prediktioner. Resultaten befanns inte vara

av tydlig praktisk nytta, med undantag av att den vektoriella

autoregressionen tydliggjorde viss regelbundenhet i variablerna

som analyserades. Vid enhetliggörandet av samplingfrekvensen

beräknades i fall av nedsampling ytterligare

deskriptiva statistika över samplingintervallen. Detta i

syfte att kvarhålla potentiellt användbar information, vilket

genom manuell inspektion kunde bekräftas vara fallet för

en av variablerna. Inkluderingen av dessa extra statistika

befanns emellertid inte ha någon tydlig positiv inverkan på

regressionsanalysen eller prediktionsförsöken.

Place, publisher, year, edition, pages
2013.
National Category
Computer Science
Identifiers
URN: urn:nbn:se:kth:diva-142348OAI: oai:DiVA.org:kth-142348DiVA: diva2:699722
Educational program
Master of Science in Engineering - Computer Science and Technology
Supervisors
Examiners
Available from: 2014-03-13 Created: 2014-02-28 Last updated: 2014-03-13Bibliographically approved

Open Access in DiVA

fulltext(1530 kB)140 downloads
File information
File name FULLTEXT01.pdfFile size 1530 kBChecksum SHA-512
c1e12ee23144555b12c9fc95d8f48b022d17cf1c3e93c49a047a94678fba39a3ec749c0758dcc0f56096fbec4fb61369e1f319dc61ce2bde6234a80ffe22f561
Type fulltextMimetype application/pdf

By organisation
School of Computer Science and Communication (CSC)
Computer Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 140 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 52 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf