Taking an existing large-scale realistic discrete event simulation model [1,2] of a mobile distributed system with more than 700 000 units we investigate the efficiency of control protocols for the software and data updates of on-board-units (OBUs) . Adding four different closed-loop control protocols to the existing simulation model we explain and investigate the technical implications of the control protocols: OBUs that allow incoming network connections can be reached at any time (Always-On or Sometimes-On). If the OBU is designed without a TCP/IP server, signaling is implemented either as polling (SometimesOn with polling) or using a dedicated GSM signaling channel (On-Demand-On). The efficiency is discussed as a combination of metrics gathered during simulation runs and compared with the OBUcontrolled update logic. These simulation results are subsequently used as input for the system development process.

We analyze existing server-side log data of a large scale automatic toll system to measure the TCP roundtrip-time (RTT) as experienced by the communication between the central system and the on-board units (OBUs) deployed for tolling heavy-goods vehicles. The RTT is estimated from passive monitoring by parsing server-side log files and aggregating fleet-wide statistics over time. Using this data we compare the characteristics of the four different types of OBU and the three GPRS (2G) networks used. We find the RTT data to be consistent with existing, smaller samples and extend the observed RTT range by an order of magnitude. The OBU types exhibit a markedly different behavior, most notably for long RTTs, and we find one of the 2G networks to 'hum' at 50 Hz and harmonics.

Taking a newly collected large data set on the TCP connection termination latency in GPRS networks we try to identify the underlying statistical distribution. The data extends the observed latencies to large time scales necessitating a heavy-tail distribution. Many distributions work well for the main body of the data. However, the heavy tail of the distribution benefits from mixing different statistical distributions. We compare several distributions and find that the double Pareto-lognormal distribution and the generalized Beta distribution of the second kind fit the data equally well.

No abstract

Simulation models of real-world distributed systems depend both on the accuracy of the underlying model and the interaction between user and system. The user interaction is typically modeled as stochastic process depending on parameters and distributions describing the actual usage. Accurate data is often not available and (manual) assumptions are necessary. Taking an existing large-scale simulation model of the German tolling system we discuss the use of a genetic optimization algorithm for calibrating the simulation model.