471.omnetpp
András Varga
Omnest Global, Inc.
andras.varga [at] omnest.com
Discrete Event Simulation
The benchmark performs discrete event simulation of a large Ethernet network. The simulation is based on the OMNeT++ discrete event simulation system (www.omnetpp.org), a generic and open simulation framework. OMNeT++'s primary application area is the simulation of communication networks, but its generic and flexible architecture allows for its use in other areas such as the simulation of IT systems, queueing networks, hardware architectures or business processes as well. The Ethernet model used in this benchmark is publicly available from the address given in the References.
For the reference workload, the simulated network models a large Ethernet campus backbone, with several smaller LANs of various sizes hanging off each backbone switch. The model contains altogether about 8000 computers (hosts), and 900 switches and hubs. It mixes all kinds of Ethernet technology: Gigabit Ethernet, 100Mb full duplex, 100Mb half duplex, 10Mb UTP, 10Mb bus ("thin Ethernet"), switched hubs, repeating hubs.
The training workload models a smaller LAN with several hubs and busses.
The model is accurate in the sense that the CSMA/CD protocol of Ethernet and the Ethernet frame are faithfully modelled. The host model contains a traffic generator which implements a generic request-response based protocol. (Higher layer protocols are not modelled in detail.) With appropriate configuration, the traffic generator can be tuned become a rough model for protocols such as SMB/CIFS (the Windows file sharing protocol), HTTP, or a database client-server protocol.
A pointer to the detailed documentation of the model can be found among the References.
The topology of the network and structure of hosts, switches and hubs are described in NED files (NED is the network description language of OMNeT++.) Operation of the Ethernet MAC, traffic generator etc. are in C++.
Several model parameters are specified in the omnetpp.ini file. These parameters include the number of switches on the backbone, the number of small, medium and large LANs on each backbone switch, the number of hosts in small, medium and large LANs as well as switch, host configuration and parameters of the traffic model. Request and reply lengths are configured in omnetpp.ini as intuniform(50,1400) and truncnormal(5000,5000) for the reference input. The volume of the traffic can most easily be controlled with the time period between sending requests; this is currently set in omnetpp.ini to exponential(0.33) (that is, average 3 requests per second) for the reference input. This already causes frames to be dropped in some of the backbone switches, so the network is a bit overloaded with the current settings.
The model generates extensive statistics. MACs (and most other model components too) write statistics into the omnetpp.sca file at the end of the simulation: number of frames sent, received, dropped, etc. These are only basic statistics; however, if all nodes were allowed to record them, omnetpp.sca would grow to about 28 megabytes. To make the output more reasonable in size, recording statistics is only enabled in a few nodes ("writeScalars" parameter in omnetpp.ini.)
The easiest way to view and plot the data in omnetpp.sca is the Scalars tool of OMNeT++ 3.0, but because the file is a simple text file, a spreadsheet program (e.g. Microsoft Excel or OpenOffice) will also do.
The model could also record output vectors (omnetpp.vec), but this feature is disabled in omnetpp.ini because the generated file can easily reach gigabyte sizes.
C++
None.
Last updated: $Date: 2011-08-16 18:23:17 -0400 (Tue, 16 Aug 2011) $