Scalable hybrid network simulation

Our methodology is based on the time-stepped hybrid simulation, initially developped in the INRIA research group TREC. .

Existing simulation approaches

There are several classes of simulation methodologies: packet level-discrete event simulation, hybrid simulation and fluid simulation.

Traditional discrete event simulation is the most accurate but it suffers from serious scalability limitations.
The fluid simulation approach scales well but fails taking into account key network engineering features such as buffering or scheduling policies, as it does not allow one to accurately simulate or analyze events, be it packet level, session level or buffer level events.

Average traffic models give orders of magnitude for capacity planning. They don't reproduce the traffic fluctuations which can be very significant on IP packet networks, especially with TCP.

Hybrid approach innovation

Our approach allows large-scale end-to-end simulation of heterogeneous (wired and wireless) networks from the terminals to the server thanks to advanced mathematical techniques that focus the simulation effort on only the relevent events, i.e. those that change the nature of traffic. It avoids simulating irrelevant packet level events which has the added advantage that it would not scale with the size of the network or the number of flows. It focuses on relevant interactions between the network and the different classes of traffic it carries through efficient packet aggregation methods per time slot. Our hybrid approach allows one to represent key engineering features such as scheduling, buffer management, and key phenomena of the current Internet such as loss synchronization, oscillations, multiple layers interaction etc.

Our technology allows one to simulate IP networks up to 10000 times larger than what can be done with existing software tools.

Scalability: The simulation time and memory requirements of NetScale™ grow linearly with traffic, whereas they grow exponentially with traditional tools with major impact on engineering efficiency. The simulation example of this curve illustrates the performance N2Nsoft's simulation technology. The computation savings depend on the actual simulation case.

It offers new possibilities and a global end-to-end vision of the network. TCP/IP traffic creates end-to-end global interactions over the network. More and more, efficient engineering of modern IP networks will require having the end-to-end view right from the start of the study.

More information can be found in:

• Data sheet:
  • HSDPA simulation data sheet.
• White papers:
  • NetScale™ fact sheet
  • Network Planning for Quality of Experience
  • End-to-end HSDPA simulation.
• A few research papers:
  • Please contact us if you are looking for research papers dealing with our technology. The following two papers are examples of mathematical analysis of the TCP behavior. They demonstrate from a theoretical standpoint why taking TCP into account can be critical for relevant network simulation.
  
  
  • [INFO03b] Baccelli, F. and Hong, D. (2003) Interaction of TCP Flows as Billiards. Proc. of INFOCOM, San Francisco, April.
    >> Network extension of the AIMD model with multi-links constraints.
  
  
  • [INFO02] Baccelli, F. and Hong, D. (2002) AIMD, Fairness and Fractal Scaling of TCP Traffic. Proc. of INFOCOM, New York, June.
    >> This paper gives a simple linear representation of N TCP connections sharing a same local loop, based on the AIMD description. In particular, it shows that the fractal property of the aggregate of TCP flows at short time scale can be partially explained by the window AIMD mechanism due to the TCP protocol. The burstiness at the packet level and the heterogeneity (different RTTs, variation of N, many routers etc.) should increase the 'irregularity' of these aggregated process. It also shows that bandwith sharing by TCP is sometimes 'very bad' (the dispersion is heavy-tailed).
  
  

---