Browse DORAS
Browse Theses
Search
Latest Additions
Creative Commons License
Except where otherwise noted, content on this site is licensed for use under a:

Evaluation of data centre networks and future directions

Sohini, Basu (2017) Evaluation of data centre networks and future directions. Master of Engineering thesis, Dublin City University.

Full text available as:

[img]
Preview
PDF (Sohini Basu MEng Thesis) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
6Mb

Abstract

Traffic forecasts predict a more than threefold increase in the global datacentre workload in coming years, caused by the increasing adoption of cloud and data-intensive applications. Consequently, there has been an unprecedented need for ultra-high throughput and minimal latency. Currently deployed hierarchical architectures using electronic packet switching technologies are costly and energy-inefficient. Very high capacity switches are required to satisfy the enormous bandwidth requirements of cloud datacentres and this limits the overall network scalability. With the maturity of photonic components, turning to optical switching in data centres is a viable option to accommodate greater bandwidth and network flexibility while potentially minimising the latency, cost and power consumption. Various DCN architectures have been proposed to date and this thesis includes a comparative analysis of such electronic and optical topologies to judge their suitability based on network performance parameters and cost/energy effectiveness, while identifying the challenges faced by recent DCN infrastructures. An analytical Layer 2 switching model is introduced that can alleviate the simulation scalability problem and evaluate the performance of the underlying DCN architecture. This model is also used to judge the variation in traffic arrival/offloading at the intermediate queueing stages and the findings are used to derive closed form expressions for traffic arrival rates and delay. The results from the simulated network demonstrate the impact of buffering and versubscription and reveal the potential bottlenecks and network design tradeoffs. TCP traffic forms the bulk of current DCN workload and so the designed network is further modified to include TCP flows generated from a realistic traffic generator for assessing the impact of Layer 4 congestion control on the DCN performance with standard TCP and datacentre specific TCP protocols (DCTCP). Optical DCN architectures mostly concentrate on core-tier switching. However, substantial energy saving is possible by introducing optics in the edge tiers. Hence, a new approach to optical switching is introduced using Optical ToR switches which can offer better delay performance than commodity switches of similiar size, while having far less power dissipation. An all-optical topology has been further outlined for the efficient implementation of the optical switch meeting the future scalability demands.

Item Type:Thesis (Master of Engineering)
Date of Award:November 2017
Refereed:No
Supervisor(s):Barry, Liam and McArdle, Conor
Subjects:Computer Science > Computer networks
Engineering > Optical communication
Engineering > Telecommunication
Computer Science > Computer engineering
Engineering > Electronic engineering
Computer Science > Computer simulation
DCU Faculties and Centres:DCU Faculties and Schools > Faculty of Engineering and Computing > School of Electronic Engineering
Research Initiatives and Centres > Research Institute for Networks and Communications Engineering (RINCE)
Use License:This item is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 3.0 License. View License
Funders:DCU
ID Code:21988
Deposited On:10 Nov 2017 15:19 by Liam Barry. Last Modified 17 Nov 2017 16:24

Download statistics

Archive Staff Only: edit this record