Custom costs balancer (CCBalancer) Floodlight module

Custom costs balancer is a Floodlight module, made to setup custom costs on the OpenFlow links cached by the Floodlight Topology module.

The Floodlight controller implements the Openflow protocol, which specifications can be found here: Openflow spec

This project depends on Floodlight, which can be found here: Floodlight project on GitHub.

It has been tested with Mininet, which can be found here: Mininet project on GitHub.

License

This sofware is licensed under the Apache License, Version 2.0.

Information can be found here: Apache License 2.0.

The Custom costs balancer module

General description

Using the standard Topology module of Floodlight, link costs are only set to 1 by the controller itself. Thus, best path are just calculated considering the minimum number of hops between two OpenFlow devices. With custom costs balancer (CCBalancer) module, users can setup via new REST API integer custom costs for each link in each direction. Thus, using external alghoritm, a user can also provide his own metric to calculate best paths. At setup time the module sets all costs to 1.

Flow rerouting

Actually, operations of rerouting are not performed, thus avoiding network balance flapping.

Installation and configuration

This project has been developed and tested on Floodlight v0.90.

The module consists of the following components:

• it.garr.ccbalancer package

• it.garr.ccbalancer.web package

• it.garr.ccbalancer.web.serializers package

• TopologyInstanceCCBalancer.java

• TopologyManagerCCBalancer.java

Eclipse

Using Eclipse, after the import of the Floodlight project:

• Copy all it.garr.ccbalancer.* packages in the project folder (outside net.floodlightcontroller)
• Copy TopologyInstanceCCBalancer.java and TopologyManagerCCBalancer.java in net.floodlightcontroller.topology
• Modify the file src/main/resources/META-INF/services/net.floodlightcontroller.core.module.IFloodlightModule
  • net.floodlightcontroller.topology.TopologyManagerCCBalancer instead of net.floodlightcontroller.topology.TopologyManager
  • add it.garr.ccbalancer.CCBalancer
• Modify the file src/main/resources/floodlight.properties
  • net.floodlightcontroller.topology.TopologyManagerCCBalancer instead of net.floodlightcontroller.topology.TopologyManager
  • add it.garr.ccbalancer.CCBalancer

Runnable file

For production environment, a jar version of the module is downloadable from the root directory of this GitHub repository.

Alternatively, it is possible to create your own ccbalancer.jar with the compiled files from this project, plus your custom IFloodlightModule file. For mode details follow the Floodlight documentation.

According to Floodlight command sintax, you can integrate the jar file to your Floodlight installation running the command:

java -cp floodlight.jar:ccbalancer.jar net.floodlightcontroller.core.Main -cf floodlight.properties

The parameters specified have the following meaning:

• floodlight.properties is the file specifying the properties for the running instance of Floodlight, it is configred to start the KHopMetric module provided with this project.

REST APIs

This project offers the following REST APIs:

• http://controller-ip:8080/wm/ccbalancer/topocosts/json
  • it supports POST actions allowing users to set via REST API custom costs for each link of the network topology, for both link directions.
  • it supports GET actions, allowing users to see actual costs, set on the links of the entire topology.

REST API tutorial

Below, an example of the REST API used to set new costs can be found:

• JSON posted to define new link costs through http://controller-ip:8080/wm/ccbalancer/topocosts/json. The following JSON is useful to set the cost of ten on links 2 - 3, 4 - 6, 1 - 2, 1 - 3.

  [
      {'src':'00:00:00:00:00:00:00:02','outPort':'2','dst':'00:00:00:00:00:00:00:03','inPort':'2','cost':'10'},
      {'src':'00:00:00:00:00:00:00:04','outPort':'4','dst':'00:00:00:00:00:00:00:06','inPort':'1','cost':'10'},
      {'src':'00:00:00:00:00:00:00:01','outPort':'1','dst':'00:00:00:00:00:00:00:02','inPort':'1','cost':'10'},
      {'src':'00:00:00:00:00:00:00:01','outPort':'2','dst':'00:00:00:00:00:00:00:03','inPort':'1','cost':'10'}
  ]
  

• JSON returned by a GET call to http://controller-ip:8080/wm/ccbalancer/topocosts/json. In this case the topology is composed by two switches connected by a link. CCBalancer returns the cost of the same link in both directions, in this case equal to 2 from switch 1 to switch 2, equal to 3 from 2 to 1.

  {
      "Link [src=00:00:00:00:00:00:00:01 outPort=2, dst=00:00:00:00:00:00:00:02, inPort=2]": 2
      "Link [src=00:00:00:00:00:00:00:02 outPort=2, dst=00:00:00:00:00:00:00:01, inPort=2]": 3,
  }
  

Utilities and examples

In the scripts folder, this projects provides some scripts that could be helpful to perform typical operations and examples:

• chktx.py is a python script that reads transmitted bits values of all interfaces using default Floodlight REST API. It saves results to a MS Excel datasheet. It requires both xlrd and xlwt python libraries to manage Ms Excel files.
• update.py is a python script that reads the transmitted bits values (if some values are saved) from a MS Excel datasheet, it computes the values to an integer cost through a sample algorithm and it sends final values to the topology module on the controller. It requires xlrd python libraries to read Ms Excel files.

More info on xlrd on https://pypi.python.org/pypi/xlrd More info on xlwt on https://pypi.python.org/pypi/xlwt

update.py algorithm

The update.py script receives the number of transmit bits from each interface activly partecipating in the network and it sends out to the controller integer costs, thus optimizing network traffic and load. Here is the sample algorithm used for computation:

cost = r * b(used)/b(tot) + s * [b(tot)-b(used)] + t * G

G = 0 if r * b(used)/b(tot) < h G = r * [[b(used) - [h * b(tot)]]/[b(tot) - [h * b(tot)]] vice versa

• where h is a percent limit of bandwidth usage, manually expressed by user in update.py
• G is a function that makes the cost degrade if the usage of the single link is higher than h
• b(used) is the used bandwidth of that link in that specific direction
• b(tot) is the total capacity of that link in that specific direction
• r, s, t are three normalization constants, also depending of the biggest capacity used in the topology

Using this the algorithm provided with the new developed Java module permit to optimize traffic load on the network, avoiding to have overloads and at the same time empty paths.

Use case

With the given code and utilities the following use case has been implemented:

• A full mesh network composed with 6 nodes have been emulated through Mininet
• For test purposes link capacity has been set to 10 Mbit
• With iperf we simulated a constant traffic between two or more sources
• With chktx.py we recorded the capacity used during the transmission
• Then, we sent another update through update.py to change link costs, depending on the traffic loads of the network.
• With another instance of iperf we started another flow between other two hosts attached to the same two switches where the first flow is going through.
• The second flow went through a different path from the first one, depending on the new link costs.

In all experiments we noted the ability of the network to adapt to different growth of traffic, avoiding overloads on congested links. Of course, using ms Excel is just a sample to easy obtain quick results and build graphs at the same time but the system can be integrated with passive monitoring systems as well (i.e. IPFIX).