Advanced

Studie av beräknings- och nätverksprestanda i 10GbE HPC-kluster

Lind, Carl (2015)
Computer Science and Engineering (BSc)
Abstract
This thesis has been performed at Saab Technologies in Gothenburg. Saab is a global defense- and security company that specializes in air-, land and marine defense. Saab in Gothenburg develops effective solutions for monitoring, warning, location and defense systems. The thesis work which has been initiated by Saab, involves development of methods and tools in order to dimension the network components in an Ethernet communicating HPC-Cluster (High performance computing). The cluster’s task is to perform signal processing for radar solutions. During the thesis work different network protocols have been evaluated in order to evaluate which protocol that proves most suitable to use for signal transmission between the radar and the cluster. It... (More)
This thesis has been performed at Saab Technologies in Gothenburg. Saab is a global defense- and security company that specializes in air-, land and marine defense. Saab in Gothenburg develops effective solutions for monitoring, warning, location and defense systems. The thesis work which has been initiated by Saab, involves development of methods and tools in order to dimension the network components in an Ethernet communicating HPC-Cluster (High performance computing). The cluster’s task is to perform signal processing for radar solutions. During the thesis work different network protocols have been evaluated in order to evaluate which protocol that proves most suitable to use for signal transmission between the radar and the cluster. It turned out that many protocols are developed or depend on the two most common protocols in the transport layer, TCP and UDP. These protocols where therefore examined more thoroughly by testing them against each other. The goal was to find out which protocol that had the upper hand by measuring their performance. This means to find out how fast both protocols can send data before unacceptable packet loss occur for the application and under those circumstances find out the packet delay as well. This cannot be done by theoretical calculations but has to be tested practically. First by testing on one computer, then between two computers and finally on the HPC-Cluster.
The result of the thesis is a number of graphs which show the two protocol’s difference in performance. (Less)
Please use this url to cite or link to this publication:
author
Lind, Carl
organization
year
type
M2 - Bachelor Degree
subject
keywords
hpc-kluster, nätverksprestanda, tcp, udp, mätning
language
Swedish
id
7472448
alternative location
http://portal.ch.lu.se/Campus.NET/Services/Publication/Export.aspx?id=2858&type=doc
date added to LUP
2015-07-01 04:39:16
date last changed
2015-07-01 04:39:16
@misc{7472448,
  abstract     = {This thesis has been performed at Saab Technologies in Gothenburg. Saab is a global defense- and security company that specializes in air-, land and marine defense. Saab in Gothenburg develops effective solutions for monitoring, warning, location and defense systems. The thesis work which has been initiated by Saab, involves development of methods and tools in order to dimension the network components in an Ethernet communicating HPC-Cluster (High performance computing). The cluster’s task is to perform signal processing for radar solutions. During the thesis work different network protocols have been evaluated in order to evaluate which protocol that proves most suitable to use for signal transmission between the radar and the cluster. It turned out that many protocols are developed or depend on the two most common protocols in the transport layer, TCP and UDP. These protocols where therefore examined more thoroughly by testing them against each other. The goal was to find out which protocol that had the upper hand by measuring their performance. This means to find out how fast both protocols can send data before unacceptable packet loss occur for the application and under those circumstances find out the packet delay as well. This cannot be done by theoretical calculations but has to be tested practically. First by testing on one computer, then between two computers and finally on the HPC-Cluster.
The result of the thesis is a number of graphs which show the two protocol’s difference in performance.},
  author       = {Lind, Carl},
  keyword      = {hpc-kluster,nätverksprestanda,tcp,udp,mätning},
  language     = {swe},
  note         = {Student Paper},
  title        = {Studie av beräknings- och nätverksprestanda i 10GbE HPC-kluster},
  year         = {2015},
}