This repository contains classes/modules that ease implementation and testing of NOC algorithms on varying topologies.
This work is done in the hope that it is found useful by scholars working in this domain.
test.py is a sample program and can be used as a quick-start.
API Documentation for all the classes is yet to be made. Appropriate comments have been made in individual class/files for the programmer's comprehension.
Invoke a topology map with
topo = topology.Topology(M,N)
where 'Topology' can be either Mesh or Torus with M and N as horizontal and vertical dimensions.
Invoking a topology only creates relevant (router/link/packet) objects. The topology has to be initialised before it starts behaving properly.
The topology has to be initialised so that the individual router elements can be linked properly to each other. This also works as a 'reset' for when all connections have to be restored to a healthy state.
topo.initialise() initialises the topology.
Once the topology is initialised, the routers can be accessed as elements of a 2D matrix.
print(topo.routers)
A nicer and more human-readable view can be obtained on the console using
topo.printTopologyMap(colour)
Setting colour to True outputs faulty routers in Red and terminal routers in Yellow. Setting it to False outputs faulty routers as empty spaces.
All routers forming the topology can be accessed using column and row indices using:
topo.routers[y][x]
Notice that the first index corresponds to the yth row and the second index to the xth column. This indexing is indeed, different from the conventional (x,y) addressing in graphs.
However, the module maintains this convention (x,y) when accepting position as a parameter or returning position as a tuple.
As stated, the only exception to this is when accessing routers directly.
The module consists of a path-finding functionality that can be accessed using
findPath(topology, source, destination), which returns the first-hit shortest path between two nodes using a modified A* approach.
The parameters source and destination are not positions but Router objects. This has been done to improve readability.
Each topology contains a heuristic function that determines the path-finding behaviour. Custom topologies need to include their unique heuristic to make use of this functionality.
showPath(topology, path) prints the map in a nice graphical view on a terminal console, with the path highlighted in Green.
Router or Link faults can be injected easily either by targeting individual routers/links or generating n random faults.
There are two ways to achieve this:
topology.injectRouterFault(topo, pos)
or
topology.injectLinkFault(topo, pos, direction)
where the parameter topo is the topology instance, pos is a tuple containing position as (x,y), and direction is the nth link to be targeted.
topo.routers[2][1].setLinkHealthList([1,1,0,1])
Here, the function parameter is a list of links' healths in the order right([0]), up([1]), left([2]), down([3]).
[1,1,0,1] means that the left link has been marked as a permanent fault.
[0,0,0,0] would mean that the router is set as faulty.
topo.routers[2][1].setLinkHealth(0,0.4)
Here, the function parameters are (direction,linkHealth).
[0,0.4] means that the link pointing to the right of the router indexed at X=1, Y=2 has a health of 0.4 (out of 1).
The topology class contains direct functions for generating 'N' random faults using
topology.injectRandomLinkFaults(topo, N, animate = True, frameDelay = 0.5) or
topology.injectRandomRouterFaults(topo, N, animate = True, frameDelay = 0.5)
animate and frameDelay are optional arguments and allow for visualising a slow injection of faults. This is not just an aesthetic addition, it also helps to see how the topology evolves with random injections, which can be very insightful for large maps with stochastically placed links. The variable names are obvious enough for defining their purpose.
The Router class is of huge importance while designing custom topologies.
Although the class is fairly flexible, a need for a derived class is possible. Remember that writing a custom Router class might be of immense help for working through a custom topology.
- Support for arbitrary number of links for each router to extend support for MoT or BFT
Consider Router-weighting for path traversal- Add FIFO buffer to router and consider Packet's size in the FIFO
Print links in topology map according to link-health