1. Wireless Sensor Network for Tracking the Traffic in INTERNET Network Routers Part 2 Supervisor:Mark Shifrin Students:Yuri Kipnis Nir Bar-Or Date:30 September 2009 Networked Software Systems Laboratory Department of Electrical Engineering Technion

2. Agenda  Terms We Use  Introduction  Project Goals  Load Balancing Algorithm  Simulation Results  Conclusions  Further Development

3. Networked Software Systems Laboratory Department of Electrical Engineering Technion Terms We Use Routers Network WSN – wireless sensor network Bellman-Ford algorithm & Shortest Path Tree (SPT) Load Balancing CWND – congestion window in TCP protocol Packet loss rate

4. Networked Software Systems Laboratory Department of Electrical Engineering Technion Solutions of improving network performance Local network solutions Network protocols WSN and Routers Network platform simulation Dynamic NS-2 platform, TCL WSN and Routers Network Load Balancing Algorithm Bellman-Ford Algorithm Introduction

5. Networked Software Systems Laboratory Department of Electrical Engineering Technion Project Goals Investigation of existed solutions Development and implementation of algorithm for improving network performance by using WSN network Running various parameterized simulations on different network topology with new algorithm Gathering simulation results, building graphs and comparing VS. regular simulation results

6. Networked Software Systems Laboratory Department of Electrical Engineering Technion The Problem What is the problem? Load Balancing

7. Networked Software Systems Laboratory Department of Electrical Engineering Technion Load Balancing Algorithm - Global Performs for all stations simultaneously pereodicly Counting the shortest path from source to destination by Bellman-Ford algorithm for SPT For each path in the SPT: if the last node is a destination of the root in routing table then calculate the load on current path and on the new path Update this path by changing an old one by new one: add new edges move edges that are not used by other flows that don’t contain the first station of old path.

8. Networked Software Systems Laboratory Department of Electrical Engineering Technion Load Balancing Algorithm – pseudo code For each node do: 1. Create SPT for node 2. newFlow list <- empty 3. Go pre-order on the SPT, for each newPath do: 3.1Is the last node in new path is a destination in RT of first node? 3.1.1No, go to 3 3.1.2Yes, Try to improve

9. Networked Software Systems Laboratory Department of Electrical Engineering Technion Load Balancing Algorithm – pseudo code Try to improve: 1.Calculate currPath between the first node of newPath and the last node of newPath 2.Get all flows that use currPath and have destination identical to newPath last node 3.Calculate the bottleneck of all these flows on old path 4.Is the current bottleneck on newPath plus calculated bottleneck lower than the total bottleneck on currPath? 4. 1 No, go to 3 on previous slide 4.2 Yes, 4.2.1 Add edges of newPath 4.2.2 Remove edges that are not used in other flows that don’t contain first node of newPath 4.2.3 go to 3 on previous slide

10. Networked Software Systems Laboratory Department of Electrical Engineering Technion Load Balancing Algorithm - example 1: Change from 0 1 2 4 to 0 5 4 1: Change from 0 1 2 3 to 0 5 3 1: Change from 1 2 4 to 1 4 1: Change from 1 2 3 to 1 3 1: Change from 5 1 2 4 to 5 4 1: Change from 5 1 2 3 to 5 3 2: Change from 0 5 4 to 0 4 2: Change from 0 5 3 to 0 3

11. Networked Software Systems Laboratory Department of Electrical Engineering Technion Flow: 0->4, QS=4, ST=0.5

12. Networked Software Systems Laboratory Department of Electrical Engineering Technion Flow: 5->3, QS=60, ST=5

13. Networked Software Systems Laboratory Department of Electrical Engineering Technion Complicated Simulation

14. Networked Software Systems Laboratory Department of Electrical Engineering Technion Flow: 0->3, QS=4, ST=0.5

15. Networked Software Systems Laboratory Department of Electrical Engineering Technion Flow 1->2, QS=4, ST=0.5

16. Networked Software Systems Laboratory Department of Electrical Engineering Technion Conclusions  Load balancing algorithm increases the network performance in networks with relatively rare number of flows and in networks with limitation in queue size and service tempo.  Load balancing algorithm increases the performance in part of flows in complicated networks that have a big number of flows and common links for different flows. This algorithm decreases a performance for other flows in complicated networks.  Networks that not suffers from the resources lack will not benefit from this algorithm in aspect of packet loss rate but in aspect of flow capacity and transfer rate.  The congestion control in wired network is possible through wireless network. Moreover it can be done by independent network like WSN by using special algorithms and protocols.

17. Networked Software Systems Laboratory Department of Electrical Engineering Technion Further Development How Load Balancing Algorithm will influence the other Internet Protocols, like UDP, BGP? How can we improve this algorithm by exchanging updated information between network stations during iteration? How can we manage the information data structures in more effective way? How can this model to be involved by industry?

18. Networked Software Systems Laboratory Department of Electrical Engineering Technion Questions ?

19. Networked Software Systems Laboratory Department of Electrical Engineering Technion Thank You ! Link to the project website: http://softlab.technion.ac.il