Virtualisation of Computer Nodes in Radar-Systems
Examensarbete för masterexamen
This thesis work has investigated the impact and implications a virtual layer could have in a transportable radar system in order to determine the feasibility of implementing such a layer in Saab’s radar systems. Today, multiple computing boards are used in the Data Processing unit of these radar systems, but in the future Saab wishestoreducethisnumber. Beforethiscanbedone,itisimportanttoknowwhat implications such achange might have. Investigating thepotential performanceloss andthebehaviourofdiﬀerentvirtuallayerscanprovideanswerstoifandwhatkind of virtual layer could be used in Saab’s systems. This has been done by measuring the performance diﬀerences between running an application on a physical machine and running the same application in a virtual layer. Our experiments show that the virtual layer itself does not have a signiﬁcant impact as long as the virtual layer operates under the same conditions as the physical machine, i.e. it has the same clock frequency the same amount of RAM etc. In addition to measuring performance, we have also made an attempt to investigate the impact the virtual layer has on the energy consumption. Our main ﬁndings are that the performance in terms of execution time is barely impacted when running CPU heavy applications in Docker containers while Docker containers have a signiﬁcant impact on I/O operations. A potential drawback from using Docker containers is that our energy measurements suggest that the container could possibly increase the energy consumption significantly. We also ﬁnd that a Xen guest could possibly have performance close to the physical machine if both machines are running on CPUs with the same clock frequency. We come to the conclusion that Xen is a feasible tool to use in order to facilitate transition from a structure consisting of multiple physical computing boards to a structure only containing one physical computing board. Our observation is that a virtual layer does not have a signiﬁcant impact on CPU performance, but the performance of I/O operations is noticeably negatively aﬀected. Energy consumption could only be measured on Docker containers and here we observe an increase in energy consumption on a process level.
Virtualisation , radar systems , Docker , Xen , performance