TU Darmstadt / ULB / TUbiblio

Observing, Predicting, and Enforcing Properties of Interactions in Data Centers

Jahnke, Patrick (2021):
Observing, Predicting, and Enforcing Properties of Interactions in Data Centers. (Publisher's Version)
Darmstadt, Technische Universität,
DOI: 10.26083/tuprints-00019073,
[Ph.D. Thesis]

Abstract

In the last decade, the rise of cloud computing combined with a tremendous increase in Internet connectivity and the need for more computational performance changed the way we conceive applications. Today, most software developers and architects design new applications as (cloud-native) distributed systems, where the demand for interactions between individual components of a given distributed system increases. Consequently, a demand-specific and efficient utilization of existing data center communication resources for distributed systems is crucial for cloud infrastructure providers. However, the main challenge in terms of efficiency is that distributed system and data center components are still very agnostic to the actual semantics of interactions. By building communication systems that are more cognizant of distributed system semantics, the performance for the corresponding components would increase tremendously. In this thesis, three fundamental challenges are tackled to improve the state-of-the-art. We show that the more precisely properties of a communication system can be observed, the more accurate the network bandwidth usage can be predicted, and the more tightly bounds on communication latency for specific distributed system (DS) interactions can be enforced.

Item Type: Ph.D. Thesis
Erschienen: 2021
Creators: Jahnke, Patrick
Status: Publisher's Version
Title: Observing, Predicting, and Enforcing Properties of Interactions in Data Centers
Language: English
Abstract:

In the last decade, the rise of cloud computing combined with a tremendous increase in Internet connectivity and the need for more computational performance changed the way we conceive applications. Today, most software developers and architects design new applications as (cloud-native) distributed systems, where the demand for interactions between individual components of a given distributed system increases. Consequently, a demand-specific and efficient utilization of existing data center communication resources for distributed systems is crucial for cloud infrastructure providers. However, the main challenge in terms of efficiency is that distributed system and data center components are still very agnostic to the actual semantics of interactions. By building communication systems that are more cognizant of distributed system semantics, the performance for the corresponding components would increase tremendously. In this thesis, three fundamental challenges are tackled to improve the state-of-the-art. We show that the more precisely properties of a communication system can be observed, the more accurate the network bandwidth usage can be predicted, and the more tightly bounds on communication latency for specific distributed system (DS) interactions can be enforced.

Place of Publication: Darmstadt
Collation: xxv, 163 Seiten
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Distributed Systems Programming
Date Deposited: 15 Jul 2021 09:02
DOI: 10.26083/tuprints-00019073
Official URL: https://tuprints.ulb.tu-darmstadt.de/19073
URN: urn:nbn:de:tuda-tuprints-190737
Referees: Mühlhäuser, Prof. Dr. Max ; Eugster, Prof. Patrick ; Fahmy, Prof. Sonia
Refereed / Verteidigung / mdl. Prüfung: 19 May 2021
Alternative Abstract:
Alternative abstract Language

Durch das Wachstum von Cloud-Infrastruktur-Anbietern in der letzten Dekade und dem damit einhergehenden Bedarf an Internet-Verbindungsbandbreite und Berechnugskapazität, hat sich die Konzipierung von Anwendungen stark verändert. Heutzutage werden die meisten Anwendungen als (Cloud native) verteilte Anwendungen entwickelt, bei denen sich der Kommunikationsaufwand zwischen den einzelnen Instanzen eines verteilten Systems erhöht. Für Cloud-Infrastruktur-Anbieter ist somit eine zielgerichtete und effiziente Ausnutzung der vorhandenen Rechenzentrennetzwerkressourcen für verteilte Anwendungen kritisch. Durch das aktuelle Nichtbeachten der Semantik von Kommunikationsmustern in verteilten Anwendungen, wird eine bestmögliche Ausnutzung schwer bis unmöglich. Durch das Erstellen eines Rechenzentrumsnetzwerks, welches sich der Semantik von verteilten Anwendungen bewusst ist, würde sich die Effizienz der entsprechenden Komponenten dramatisch steigern. In dieser Doktorarbeit werden neue Ansätze in drei wichtigen Bereichen vorgestellt, die den Status quo verbessern. Es wird gezeigt, dass je präziser Eigenschaften eines Rechen- zentrumsnetzwerks beobachtet werden, genaue Vorhersagen zur Bandbreitennutzung von Kommunikationssystemen möglich sind und kleine strikte obere Schranken der Latzenz für spezifische Interaktionen in verteilten Systemen bestimmt werden können.

German
Export:
Suche nach Titel in: TUfind oder in Google
Send an inquiry Send an inquiry

Options (only for editors)
Show editorial Details Show editorial Details