Site maintenance: Ilka Miloucheva

Efficient inter-domain simulator based on time series data

• Efficiency through simulation on aggregate level in a large-scale discrete time space - simulation time depends on the length of the time series, the number of border routers and the complexity of the border router models. The aggregation level of the time series can be chosen freely. More detailed (longer) time series result in more accurate values but the simulation time will increase. More aggregated time series result in aggregate values in competitive run time.
• The simulated network consists of network domains. Each domain is represented by its border routers. Time series data scheduling occurs at border routers in to distinct direction: input and output -- this is meant in scope of the domain and not of the router. On the domain level the simulator assumes that the input and output traffic has no effect on each other.

Simulation architecture that works on time series

• Time series simulator evaluates the QoS state of the network in terms of throughput, drops, delay and jitter. Each border router processes the time series that represents the input traffic and produces time series that are forwarded to the successive border routers that handle them as output traffic (see Illustration). (Again, input and output is meant to be domain input and output.) In addition, the border routers compute QoS state information on the same aggregation level as the traffic is described in the time series.
• Since the simulator is based on aggregate load information the QoS values represent an average for the traffic as a whole. The more detailed the load information is the more accurate the QoS values will be with an increasing run time. However to keep that assumption of the simulator real that the domain input and output traffic has no effect on each other, one should keep the aggregation level of the load description on a certain level.
  
  Illustration - Forwarding of time series among border routers
  

Applications

• Analysis the effect of additional traffic on the QoS state of the network. Normally, the simulator evaluates the QoS state of the network based in the load information (time series of traffic). This information could be a result of a measurement process somewhere in the network. By increasing (decreasing) the values in this traffic description one can simulate additional traffic in the network, then comparing the QoS values of the original and the added-traffic scenario the changes reveal.
• Analysis the effect of rerouting a part of the traffic on the QoS state of the network. Routing information in the simulator is stored in form of traffic distributions. As a border router processes the input traffic it generates time series for the successive border routers (see Illustration 1). So, the traffic that enters to the domain should be distributed among the other border routers of the domain. This distribution information acts as aggregate routing information in the simulator and normally retrieved from measurement data in a real life scenario. Changing the weights in the distribution yields in new traffic conditions as some part of the traffic would be rerouted.