Experimental Assessment of Benchmark-oriented Network Traffic Generators

Abstract

The burstiness of network traffic has a profound impact on the performance on many network protocols in areas such as, congestion control (eg. TCP), multiple-access (eg.CSMA/CA), routing (eg. BGP), switching and multiplexing in general. Network measurement performance tools, have been widely used to exploit vulnerabilities, monitoring reports and testing newly, under development protocols. Network Traffic Generators (NTG), are capable of tracing burden from real-life traffic and replicate it offline, generating load under predefined traffic profiles such as, CBR, VBR and traffic according to well known probability distributions (Poisson process and Poissonianarrival). Though, it is not clear how accurately these tools are performing, as little study has been done. A traffic generator is required to satisfy hard real time requirements; time sensitive applications (eg. VoIP, Video), replication/emulation could lead to different than the expected experimental outcome, mainly because of the intrinsic limitations of the PC architecture and NTG’s design. Processes are competing each other for CPU attention,rising up uncertainties possibility and higher kernel overhead. The accuracy of timers , the network socket family and the policy of process scheduling should be consider as main sources of fluctuations and eventually bursty traffic generation. In this work, we use several well known statistical tools for capturing anomalies in frame generation process. Developers should take into account the rising limitations for better NTG design and implementation.