LOGICAL TOPOLOGY DESIGN Practical tools to configure networks Guido. A. Gavilanes February, 2010 1
Introduction to LTD " Design a topology for specific requirements " A service provider must optimize its costs " Lightpaths can be optimized regarding many constraints " Forward traffic in the electronic domain to nodes is easy but expensive, (use of ADM resources, causes delay) " But also transmitters cost, so not a fully connected solution is feasable " You can also minimize delay by having few hops in your network " Many optimizations can be performed, if delay is not important, perhaps you can minimize the number of links " But this optimization can not be solved deterministically just by equations, and time to arrive to a solution can not be determined precisely 2
Network Graphs Tool " In such a situation, a service provider must find a way to perform that optimization " It requires memory structures to characterize a network topology (lightpaths,bandwidths,direction) and traffic requirements " It must calculate routes at all nodes and verify topology for isolated nodes " These and many additional functions are available in this tool to the algorithm designer (you!). 3
External tools " Open http://www.telematica.polito.it/mellia/corsi/07-08/reti_ottic and download graph libraries " Install Dev c++ IDE (integrated development environment + compiler) " download graph.zip and graph-rwa.zip " open graph (decompressed) and open the file.dev file with dev c++ " graph library is written in c 4
About Random Number Generators " Some of our algorithms to build networks will require random numbers " A simple generator: X i+1 =(a X i + c)mod m " X 0 is the seed " c = increment (c=0 congruent multiplicative generator) " do the experiment with m = 11 a=7 c=0 X 0 =1 " All generators will have a period " Seed is relevant to be able to generate sets of random numbers and repeat experiments 5
LTD solution " Remember that the optimal solution almost will never arrive " What we can do is to implement some heuristic algorithms # Optimize number of links necesary # or Minimize mean number of hops source-destination # or minimize maximum lightpath congestion max(f ij ) # perhaps constraining a maximum number of links per node δ i O δi I 6
Hands on work " Look at the main example " Build some simple topologies # Full Mesh # Random graph # Deterministic Ring # Deterministic Bidirectional Ring * # Manhattan Network * # Random Ring " Recommended exercises before the report: # Build a program that starting from a given number of nodes, uses as less links as possible: build the simplest connected network), this would eventually converge to a unidirectional ring. 7
Excercises " Try to build these topologies: Manhattan network, Bidirectional Ring 0 1 10 11 2 4 3 5 12 14 13 15 0 1 5 6 7 16 17 2 4 8 9 18 19 LTD 3 LTD 8
Visualization tool " Download dot from www.graphviz.org " Download graph2.zip from www.tlc-networks.polito.it " Take a look at the print-structures files and observe the last output of the program " export your plots in jpg 9
A practical more complex LTD example " An example of final homework " A Tabusearch Implementation " with tests and results on performance 10
Routing and Wavelength Assignment " Given Physical topology " Given LTD requests " find wavelengths associated to the requests 11
Thanks! P1! P2! P3 Thank you for your atention! 12
Project Thanks! P1! P2! P3 Project 1 Project 2 " Given a random traffic matrix (generated only once, see traffic matrix example) " Given a δ I/O for all nodes " Solve the LTD to minimize congestion over links using operative research algorithms, local search, heuristics or both; implement two heuristics (or two variants of your algorithms). " document the performance of your algorithm plotting convergence time Vs. scenario for both variants. " document dependency of maximum congestion Vs. N or σ (examine your results and decide which). " Given a traffic matrix (generated only once, see traffic matrix example) " Given a maximum value od δ I/O for all nodes " Solve the LTD minimizing: congestion using operative research algorithms, local search, heuristics or both. " Solve the LDT also minimizing: the amount of transmitters+receivers turned used (and without caring about congestion). " Document the mean amount of transmitters/receivers that you can turn off when changing the optimization target (congestion- amount of tx on) for different values of N. Report also the difference in congestion. Report 1 scenario with both optimizations graphically. Document also performance (in free format). 13
Project (2) Thanks! P1! P2! P3 Project 3 " Given two different traffic matrixes (generated only once, see traffic matrix example) one random and another in which the node with the highest traffic out is lowered down by 50 percent, the second by 25 percent and all others by 20 percent. " Given a maximum value od δ I/O for all nodes " Solve the LTD maximizing f ij (lightpath usage) in both cases using operative research algorithms, local search, heuristics or both. Remember that the maximum lightpath usage cannot be higher that the link capacity. " Document the mean amount of transmitters/receivers that you can turn off when using matrix 2 for different values of N. Report also the difference in congestion. Report 1 scenarios with both matrixes graphically. Run many realizations (remember to change the seeds) and report mean values in all the cases 14
Project (3) Thanks! P1! P2! P3 Project 4 " Given a traffic matrix (generated only once, see traffic matrix example) with 2 nodes not producing/receiving traffic " Assume a physical topology G(n,p) with probability. " Assuming that the power consumed per lightpath for transmitters and oxcs is related as α lp = P tx/lightpath P oxc/lightpath and that α SW = P electsw ON P oxc ON " " Solve the LTD to minimize total network power consumption using operative research algorithms, local search, heuristics or both; Note that you must take into account where the lightpaths are routed in the physical topology. (don t care about the wavelength assignment, and you don t have to change routing). " Report the power optimized for different values of probability p of the physical topology. " Report also the time required by the algorithm to arrive to an acceptable solution. " Use a δ I/O by node as you prefer, and vary α lp and α SW in [2, 4, 8, 16]. " Report your observations freely If you build an iterative algorithm, (as a suggestion) plot the target function value on each iteration, in order to verify convergence to more optimized values on each iteration. 15