Detailed Simulation of Large-Scale Wireless Networks
3 likes516 views
The document discusses a detailed simulation approach for large-scale wireless networks, emphasizing the limitations of monolithic simulators and advocating for parallel and distributed simulation techniques to enhance scalability and efficiency. It outlines proposed tools and mechanisms, such as the ARTÌS middleware, GAIA+ for load balancing, and WIFRA for implementing wireless models, alongside experimental evaluations demonstrating the feasibility and performance improvements of these approaches. Future work aims to further develop these frameworks and add more protocols to enhance simulation capabilities.
Detailed Simulation of Large-Scale Wireless Networks
- 1. Gabriele D’Angelo joint work with Michele Bracuto University of Bologna Department of Computer Science DS-RT 2007 – Crete Island (Greece) Detailed Simulation of Large-Scale Wireless Networks
- 2. Presentation outline Basic assumptions and goals Parallel and Distributed Simulation (PADS) Synchronization and load balancing Tools and mechanisms: ARTÌS, GAIA+, WiFra Experimental evaluation Conclusions and future work
- 3. Basic assumptions Many of the systems of interest are composed by a very large number of entities, with a really dynamic nature and evolution Wireless networks, under the simulation viewpoint, have very strict requirements in terms of level of detail Networks composed of hundreds of thousands up to millions of nodes will be widely diffused in the next years Monolithic simulators are unable to fulfill such scalability requirements due to memory constrains and excessive amount of time required to obtain the results
- 4. Basic assumptions Parallel And Distributed Simulation can be used to aggregate memory and computational resources A set of interconnected Physical Execution Units (PEUs) is responsible to manage and evolve the simulation state This approach is not free from drawbacks, the distributed execution architecture needs appropriate mechanisms for: Synchronization (Communication) Load-balancing
- 5. Synchronization Under the synchronization viewpoint, the detailed (and efficient) simulation of wireless networks is a complex problem The Medium Access Control (MAC) protocols commonly used in wireless network requires very fine-grained models to represent the state of the shared medium and the behavior of wireless devices In example, in the case of 802.11 DCF the Short Interframe Space (SIF) is set to 10 μ s. This potentially means a huge number of synchronizations in the distributed execution architecture
- 6. Goals Main goals of this work, demonstrate that: Large scale wireless networks (200.000 node) can be efficiently simulated following an appropriate distributed simulation approach Specifically tailored techniques can improve the communication efficiency and therefore increase the simulation speed Very large-scale wireless networks (1.000.0000 nodes) can be simulated using clusters composed of Commercial-Off-The-Shelf (COTS) hardware
- 7. Distributed simulation: definitions The distributed simulation is run by a set of Physical Execution Units (PEUs) Each PEU is responsible to manage the execution of a Logical Process (LP) Each LP, simply works as a container: it manages the evolution of a set of Simulated Model Entities (SMEs)
- 8. Synchronization + Load Balancing As said before, synchronization (more generally the communication overhead) and load-balancing are two of the main problems to address in PADS The reduction of the communication overhead and the computation load-balancing can be seen as different aspects of the same problem and therefore should be addressed together A single mechanism should manage both of them The presence of heterogeneous hardware (e.g. different CPU models) can be seen as a special load balancing requirement
- 9. Migration-based approach We propose a migration-based approach: every entity in the simulation can be dynamically relocated (migrated), moving from a source LP to a new destination LP Clustering in the same LP the highly interacting Simulated Mobile Entities (SMEs) it is possible to reduce the costly inter -LP communication and conversely increasing the rate of low cost intra -LP communication
- 10. Load-balancing This approach can be also used to improve the computation load-balancing of the execution architecture The synchronization points in the distributed architecture can be used to tag each LP as “ fast ” or “ slow ” A LP is “slow” if: Its PEU (i.e. CPU) is overloaded Its communication network has a higher delay with respect to other parts of the execution architecture In both cases, to speed up the simulation, a solution is to migrate some entities from the slow LPs to faster LPs, therefore reducing the imbalance
- 11. Load-balacing This description of the mechanism is very high level, the “real world” implementation has to take in account many subtle details, some of them: Each reallocation has a cost that depends on many factors The algorithm has to be “fully distributed” (without any centralization point) and therefore without any “global vision” of the distributed system The mechanism has to quickly react to internal (i.e. the creation of new entities) and external events (i.e. a burst of CPU or network load), without introducing oscillatory behaviors
- 12. ART ÌS: parallel and distributed simulation middleware The (Advanced RTI System) ARTÌS is a middleware for Parallel and Distributed Simulation Adaptively adjusts the communication behavior with respect to network technology (that is, adaptively selects Shared-Memory, R-UDP, SCTP, TCP etc. to improve the communication performances) Homepage: http://pads.cs.unibo.it
- 13. GAIA+ : communication and computation load-balancing GAIA+ framework is responsible to adaptively manage the migration of the simulated entities It is composed by a set of tightly coupled heuristics, that take care of the computation and communication load-balancing It provides an easy to use abstraction level for the definition of new simulation models
- 14. WiFra : Wireless Framework The Wireless Framework (WiFra) is a collection of wireless models In this case we have implemented a IEEE 802.11 DCF MAC Layer protocol and on top of it a very simple info-mobility application Each entity defined in WiFra can be automatically migrated by GAIA+, depending on the defined heuristics and the tuning parameters
- 15. Experimental evaluation Wireless model parameters Simulation time-step 10 μs (802.11 SIFS) MAC layer IEEE 802.11 DCF Packet size 1024 bytes Packet rate 4 pkt/s Propagation model Transmission range Free space propagation 250 meters Simulated area Variable size Fixed density of nodes Nominal channel bit rate 2 Mbps Mobility model Random WayPoint (RWP) Simulated devices (SMHs) 200.000, 1.000.000
- 16. Experimental evaluation: distributed simulation environment 200.000 SMHs. 1 up to 8 PEUs: Dual Core Intel Pentium IV CPU 3.0 GHz with 1GB of RAM, interconnected by Fast Ethernet
- 17. Experimental evaluation: GAIA+ ON/OFF
- 18. Experimental evaluation: massively populated scenario Very large-scale wireless network composed of 1.000.000 nodes Execution architecture: 32 PEUs: Dual Core Intel Pentium IV CPU 3 GHz 1GB RAM, interconnected by Fast Ethernet Each PEU initially allocates 32.250 SMHs During the simulation the GAIA+ mechanism triggers re-allocations to adapt the load of each PEU to the performance of the hardware and background load (e.g. other running tasks)
- 19. Experimental evaluation: massively populated scenario
- 20. Experimental evaluation: massively populated scenario The gain in terms of performance is up to 21% More complex user level protocols would increase this gain
- 21. Conclusions and future work Distributed simulation is a feasible approach for the detailed (fine-grained) simulation of large-scale wireless networks An approach based on adaptive migration of the simulated entities can both enhance the communication and computation load balancing, therefore increasing the simulation speed-up We plan to add more protocol models to WiFra Further develop the GAIA+ framework to adaptively activate more CPUs (on-demand) and to reduce the number of used CPUs when are not necessary, therefore reducing the synchronization overhead
- 22. Gabriele D’Angelo joint work with Michele Bracuto University of Bologna Department of Computer Science DS-RT 2007 – Crete Island (Greece) Detailed Simulation of Large-Scale Wireless Networks