Summary: Merge of discrete packet simulation with fluid simulation, by transforming the packets into fluid rates.
The network is splitted into a fluid central part and different discrete parts. The discrete packets enter the fluid network and exit after some delay D and loss probability Lp.
Discrete packets are sampled every "smoothing period" S, which the fluid solver uses to calculate the values (D, Lp) for next T+S seconds.
5 Experiment show very good accuracy compared to 100% discrete simulation, and a speedup of O(10).
The fluid model used in the one in [6], which estimates TCP congestiion window in AQM (ej: Random Early Discard) networks).
Extended [6] to calculate "traffic propagation" (arrival and departed rates at each station) and "Queue Evolution" (queue size and each station)
The discrete packets can be any protocol (TCP and UDP in the experiments)


Ideas para mejorar nosotros con DEVS:
- Mismo simulador y solver (en el paper son separados)
- Aca cada parte de la red es o continua o discreta. Nuestra idea es hacer cada nodo hibrido. De esta forma no habria un transformacion de paquete a flujo como en el paper.
- Agregar prioridades en el ruteo (parece facil en la ecuacion 3). Util para el modelo de TDAQ
- Reusar las "traffic propagation" and "queue evolution" equations
- Ver de sacar la parte de AQM de la ecuaciones.
- [6] asume que los clientes TCP tienen constantemente datos para enviar (ej: video, audio). Ver si se puede reemplazar por otras ecuaciones para clientes no TCP (sin descartes?, con rafagas?)
-


- Next reads:
[6]"topology-aware fluid simulation", Y. Liu, F. L. Presti, V. Misra, D. Towsley, and Y. Gu, “Fluid mod-
els and solutions for large-scale ip networks,” in Proceedings of
ACM/SIGMETRICS ’2003, June 2003.
[8] "model network traffic as fluid flows and generate discrete events for packet losses within the network"
[9] "outlines the design of a fluid-oriented hybrid simulator admitting both packet flows and fluid flows."

-- MatiasAlejandroBonaventura - 2016-05-31