Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GPU AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi and B.H.V. Topping
Performance evaluation of massively parallel communication sessions
Z. Gál1, I. Varga1, T. Tajti2, G. Kocsis1, Z. Langmajer1, M. Kosa1 and J. Panovics1
1Faculty of Informatics, University of Debrecen, Hungary
Z. Gál, I. Varga, T. Tajti, G. Kocsis, Z. Langmajer, M. Kosa and J. Panovics, "Performance evaluation of massively parallel communication sessions", in P. Iványi, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Parallel, Distributed, GPU and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 34, 2019. doi:10.4203/ccp.112.34
Keywords: high speed networking, parallel communication, Internet, congestion control, traffic engineering, statistical analysis, scale independence.
Main issues of the Big Data processing includes not just computation or storing huge amount of bytes but high speed transmission of these data between different nodes of the ICT system, as well. Best effort based datagram delivery of the protocol data unit streams provides usable time critical services just in networks having minimum bandwidth in the scale of n*10 Mbps. Although IntServ and DiffServ QoS mechanisms make possible for time critical data flows to be forwarded in reasonable conditions, high speed transmission of the big data in LAN/WAN environments remains hot topic. Different implementations of the TCP congestion control mechanism with divisive efficiency of the transmission speed were developed by research institutes, standardisation institutes and ICT companies in the last decades. The QoS strategies applied in LAN environment are weakly usable in wide area data networks producing low usage efficiency of the communication path traversing different ISPs. Comparison and analysis of the representing high speed communication mechanisms existing today makes possible to set configuration for best transmission performance. In our performance evaluation framework we developed an measurement tool based on iperf and tcpdump software to capture data flows of TCP and UDP sessions. Classification aspects of the measurement cases were: homogeneity of the traffics (just TCP, just UDP, heterogeneous TCP and UDP), number of parallel communication sessions (1...100) and type of the TCP congestion control algorithm (16 different implementations). More than four hundred traffic scenarios were executed in a dump-bell topology containing routers and Linux machines. Statistical analysis methods were used to evaluate effects of the aspects mentioned above in the wireline local and metropolitan area networks. All the research results of this framework serve as design and development consideration of a high speed reliable file transfer application based on UDP protocol.
purchase the full-text of this paper (price £22)