Statistics
| Branch: | Revision:

mobicen / documents / old / direzioni / direzioni.md @ 9110387b

History | View | Annotate | Download (8.21 KB)

 1 ---  title: "How to study Pop-Routing with mobility"  subtitle: "List of possible experiments with PROs and CONs"  author: []  date: \today  #subject: "Network algorithms"  #keywords: [Markdown, Example]  titlepage: true  #titlepage-color: "06386e"  #titlepage-text-color: "FFFFFF"  #titlepage-rule-color: "FFFFFF"  #titlepage-rule-height: 1  header-includes:   - \usepackage{xspace}   - \usepackage{amsmath}   - \usepackage{amssymb}   - \usepackage{cleveref}   - \usepackage{hyperref}   - \hypersetup{colorlinks=true, linkcolor=blue, filecolor=magenta, urlcolor=cyan}  ...  # Goal  Verify and quantify the **performance gain** we may have introducing POP-routing in mobile wireless networks.  # Performance metric  I believe these could be the possible  performance indicators worth to be studied in function of **mobility-characteristic parameters**, this  to compare a mobile POP-network with a non-POP one:  1. Average Non working Path per second  2. Throughput  3. Logical-to-physical graph distance  4. Convergence speed of centrality  ## How to compute them  1. *Average Non working Path per second*  - Periodic (and wishfully synchronized) sample of all RTs of all nodes, sample and record also the physical graph with a timestamp.  - ==> Offline navigation of RTs, indexed and grouped per timestamp, to count how many routes installed by nodes are physically broken.  EXPECTED RESULT:  With POP we have fewer broken-paths per second  2. *Throughput*  - Let all nodes generate continuously broadcast traffic  - Measure received/lost traffic over a whole simulation timeframe  EXPECTED RESULT:  With POP we have fewer losses / greater throughput  3. *Logical-to-physical graph distance*  - Sample periodically RTs of nodes and the physical network topology  - Build the logical graph mantained by nodes at the routing layer  - Compute some graph-distance, e.g.: [Graph edit distance](https://en.wikipedia.org/wiki/Graph_edit_distance)  EXPECTED RESULT:  With POP the logical graph is on average more similar to the physical one.  If not, at least the graph-distance of the core of both logical and physical graphs are more similar if we use POP.  4. *Convergence speed of centrality*  - Classic convergence study of centrality computation,  this time in function of "mobility-characteristic parameters"  EXPECTED RESULT:  For not extremely fast networks the computation converges and is stable enough to claim that, for some time after,  POP-routing could be used bringing all the advantages already known for static networks.  # Simulation Approach  Candidate tools: custom-Python DES, Omnet...  Many CHALLENGES:  0. *Model Mobility*   a. A nice [python implementation of mobility models](https://github.com/panisson/pymobility)  1. *Model the routing protocol*   a. neighbour discovery and management   b. route update propagation (or, in case of LS, TC propagation)  2. *Implement POP-routing on top of the modelled Routing Protocol*   a. Implement Load Centrality distributed algortihm   b. or, for LS, Brandes/Prince/[dynamic BTW](http://delivery.acm.org/10.1145/2940000/2939770/p1145-riondato.pdf?ip=193.205.210.74&id=2939770&acc=CHORUS&key=296E2ED678667973%2E532136EDD1F8E584%2E4D4702B0C3E38B35%2E6D218144511F3437&__acm__=1548087694_5659b59dca6726188fcc43e73c8d95aa)   c. Implement dynamic timer adjustments on top of centrality computation  3. *Model radio channels*   - [A complicated but python tool](http://pylayers.github.io/pylayers/)   - or some basic self-implemented workaround. Unit-disk model? Or something little better: some prob of failure distribution and some pathloss dependent on link-length   - Or nothing at all, perfect and instantaneous links!  Other candidate tools:  - \url{https://www.gnuradio.org/doc/doxygen/page_channels.html}  - \url{https://github.com/intrig-unicamp/mininet-wifi}  - \url{https://github.com/bcopeland/wmediumd}  ### Considerations on Omnet  Oment for sure provides some mobility libraries and advanced radio channels. The challenges adopting Omnet would be the integration of  mobility and radio libaries with a routing protocol (among the many already implemented in Omnet) customized to implement POP-routing.  Summing up, I am not an expert of Oment but I should somehow implement POP-routing inside Oment, and mix many "subprojects" of Oment to be able to  setup an appropriate enviroments for experiments  ## PROs and CONs  The approximations introduced while modelling mobility, the routing protocol and wireless channels can't be neglected. Metrics like AVG broken Paths per second or Throughput are not suited for a simulation study. The other 2 metrics (graph distance and convergence speed [measured in "Virtual time"]), could be instead studied in the flexible environment of a python-simulator, but are less informative metrics. The mobility patterns offered by the python-tool already identified seems to offer very good models.  # Emulation Approach  The candidate tools are NePa, Mininet-wifi ...  CHALLENGES:  0. How to embed mobility in emulators?  1. OLSR + Prince or Babel + LC? For the moment my customBabel only computes LC. LC-Dissemination and timer-tuning still need to be implemented  2. Radio modelling? Emulated WiFi drivers available? ...  ## PROs and CONs  If we manage to setup things then also the two main metrics could be studied. However, the setup is really hard! No known emulators offer Mobility + radio channels together and, moreover, customBabel is not ready for emulation.