glanf VNF dockerfiles

This repository contains NFs implemented by the authors of the GLANF framework.

Installation

This setup requires sudo permissions and also makes network level changes. Thus it is best to experiment with it in a Linux VM or use the testing docker image within which docker is installed.

---

Pre-req

1. If running glanf in host or VM then you need the pipework tool

   • You can get it here
   • It is a shell script named pipework -Just copy this file to /usr/bin

2. Must have docker installed

3. Must have python package flask-restful installed to run the test webapp

   • You can install it using pip : pip install flask-restful

---

Build

• Build all VNF docker-images using the buildAll.sh script.

  • run docker images to view the image names

• This script will create the following aliases in your .bashrc file:

  • glanf_start, glanf_stop, glanf_clean, glanf_reset
  • each of these commands simply run this glanf script.

---

Running the glanf script

- glanf_start <BASE_IMAGE> <INTERMEDIARY_VNF_1_IMAGE> <INTERMEDIARY_VNF_2_IMAGE> ....
- glanf_stop (stops all the running containers)  
- glanf_reset (stops any running containers, cleans the connections, cleans the OVS bridge)

Always use glanf_reset between subsequent calls to glanf_start to get a clean state.

---

glanf_start

Ex: glanf_start glanf/base glanf/wire

• The above example has used the glanf/base as the base-image and has one middlebox container from the image glanf/wire

• The setup is always as above. The source and destination containers are from the glanf/base image whilst the middle box is from the glanf/wire image.

• In the meantime the containers are connected to the ovs-bridge and flow rules are setup to forward traffic between the source and destination containers via the intermediary MIDDLEBOX containers.

• There could be more than one middlebox images passed to glanf_start, in which case all would be chained between the source and destination.

---

VNF - firewall

---

loadsim - VNF

This NF behaves as glanf/wire does, but with additional delay added and load performed for each packet which passes through the device. This is meant to simulate devices which have inherent baseline latency which they add and which perform processing per packet.

The amount of "processing" performed per packet scales exponentially according to a load factor, which can be tweaked by configuring the LOAD_FACTOR environment variable (which can be a floating point value). The amount of processing performed is equal to LOAD_FACTOR * ( 2 ** LOAD_FACTOR) which has the nice property that it scales from 0 to a large number very quickly.

The delay is added to outgoing packets on both interfaces.

An example invocation is as follows:

docker run -itd --cap-add=NET_ADMIN --env="LOAD_FACTOR=8.0" --env="DELAY=2ms" glanf/loadsim