Policy-controlled Video Streaming (pcvs) with APEX

Introduction

This module contains several demos for Policy-controlled Video Streaming (PCVS). Each demo defines a policy using AVRO and Javascript (or other scripting languages for the policy logic). To run the demo, a vanilla Ubuntu server with some extra software packages is required:

  • Mininet as network simulator

  • Floodlight as SDN controller

  • Kafka as messaging system

  • Zookeeper for Kafka configuration

  • APEX for policy control

Install Ubuntu Server and SW

Install Demo

Requirements:

  • Ubuntu server: 1.4 GB

  • Ubuntu with Xubuntu Desktop, git, Firefox: 2.3 GB

  • Ubuntu with all, system updated: 3 GB

  • With ZK, Kafka, VLC, Mininet, Floodlight, Python: 4.4 GB

  • APEX Build (M2 and built): M2 ~ 2 GB, APEX ~3.5 GB

  • APEX install (not build locally): ~ 300 MB

On a Ubuntu OS (install a stable or LTS server first)

# pre for Ubuntu, tools and X
sudo apt-get  -y install --no-install-recommends software-properties-common
sudo apt-get  -y install --no-install-recommends build-essential
sudo apt-get  -y install --no-install-recommends git
sudo aptitude -y install --no-install-recommends xubuntu-desktop
sudo apt-get  -y install --no-install-recommends firefox


# install Java
sudo add-apt-repository ppa:webupd8team/java
sudo apt-get update
sudo apt-get -y install --no-install-recommends oracle-java8-installer
java -version


# reboot system, run system update, then continue

# if VBox additions are needed, install and reboot
sudo (cd /usr/local/share; wget https://www.virtualbox.org/download/testcase/VBoxGuestAdditions_5.2.7-120528.iso)
sudo mount /usr/local/share/VBoxGuestAdditions_5.2.7-120528.iso /media/cdrom
sudo (cd /media/cdrom;VBoxLinuxAdditions.run)


# update apt-get DB
sudo apt-get update

# if APEX is build from source, install maven and rpm
sudo apt-get install maven rpm

# install ZooKeeper
sudo apt-get install zookeeperd

# install Kafka
(cd /tmp;wget http://ftp.heanet.ie/mirrors/www.apache.org/dist/kafka/1.0.0/kafka_2.12-1.0.0.tgz --show-progress)
sudo mkdir /opt/Kafka
sudo tar -xvf /tmp/kafka_2.12-1.0.0.tgz -C /opt/Kafka/

# install mininet
cd /usr/local/src
sudo git clone https://github.com/mininet/mininet.git
(cd mininet;util/install.sh -a)

# install floodlight, requires ant
sudo apt-get install ant
cd /usr/local/src
sudo wget --no-check-certificate https://github.com/floodlight/floodlight/archive/master.zip
sudo unzip master.zip
cd floodlight-master
sudo ant
sudo mkdir /var/lib/floodlight
sudo chmod 777 /var/lib/floodlight

# install python pip
sudo apt-get install python-pip

# install kafka-python (need newer version from github)
cd /usr/local/src
sudo git clone https://github.com/dpkp/kafka-python
sudo pip install ./kafka-python

# install vlc
sudo apt-get install vlc

Install APEX either from source or from a distribution package. See the APEX documentation for details. We assume that APEX is installed in /opt/ericsson/apex/apex

Copy the LinkMonitor file to Kafka-Python

sudo cp /opt/ericsson/apex/apex/examples/scripts/pcvs/vpnsla/LinkMonitor.py /usr/local/src/kafka-python

Change the Logback configuration in APEX to logic logging

(cd /opt/ericsson/apex/apex/etc; sudo cp logback-logic.xml logback.xml)

Get the Demo Video

sudo mkdir /usr/local/src/videos

Standard 720p (recommended)

(cd /usr/local/src/videos; sudo curl -o big_buck_bunny_480p_surround.avi http://download.blender.org/peach/bigbuckbunny_movies/big_buck_bunny_480p_surround-fix.avi)

Full HD video

(cd videos; sudo curl -o bbb_sunflower_1080p_60fps_normal.mp4 http://distribution.bbb3d.renderfarming.net/video/mp4/bbb_sunflower_1080p_60fps_normal.mp4)

VPN SLA Demo

This demo uses a network with several central office and core switches, over which two VPNs are run. A customer A has two location A1 and A2 and a VPN between them. A customer B has two location B1 and B2 and a VPN between them.

VPN SLA Architecture

The architecture above shows the scenario. The components are realized in this demo as follows:

  • CEP / Analytics - a simple Python script taking events from Kafka and sending them to APEX

  • APEX / Policy - the APEX engine running the VPA SLA policy

  • Controller - A vanilla Floodlight controller taking events from the Link Monitor and configuring Mininet

  • Network - A network created using Mininet

The demo requires to start some software (detailed below). To show actual video streams, we use VLC. If you do not want to show video streams, but only the policy, skip the VLC section.

All shown scripts are available in a full APEX installation in $APEX_HOME/examples/scripts/pcvs/vpnsla.

Start all Software

Create environment variables in a file, say env.sh. In each new Xterm

  • Source these environment settings, e.g. . ./env.sh

  • Run the commands below as root (sudo per command or sudo -i for interactive mode as shown below)

#!/usr/bin/env bash

export src_dir=/usr/local/src
export APEX_HOME=/opt/ericsson/apex/apex
export APEX_USER=apexuser

In a new Xterm, start Floodlight

sudo -i
. ./env.sh
cd $src_dir/floodlight-master && java -jar target/floodlight.jar

In a new Xterm start Mininet

sudo -i
. ./env.sh
mn -c && python $APEX_HOME/examples/scripts/pcvs/vpnsla/MininetTopology.py

In a new Xterm, start Kafka

sudo -i
. ./env.sh
/opt/Kafka/kafka_2.12-1.0.0/bin/kafka-server-start.sh /opt/Kafka/kafka_2.12-1.0.0/config/server.properties

In a new Xerm start APEX with the Kafka configuration for this demo

cd $APEX_HOME
./bin/apexApps.sh engine -c examples/config/pcvs/vpnsla/kafka2kafka.json

In a new Xterm start the Link Monitor. The Link Monitor has a 30 second sleep to slow down the demonstration. So the first action of it comes 30 seconds after start. Every new action in 30 second intervals.

sudo -i
. ./env.sh
cd $src_dir
xterm -hold -e 'python3 $src_dir/kafka-python/LinkMonitor.py' &

Now all software should be started and the demo is running. The Link Monitor will send link up events, picked up by APEX which triggers the policy. Since there is no problem, the policy will do nothing.

Create 2 Video Streams with VLC

In the Mininet console, type xterm A1 A2 and xterm B1 B2 to open terminals on these nodes.

A2 and B2 are the receiving nodes. In these terminals, run vlc-wrapper. In each opened VLC window do

  • Click Media → Open Network Stream

  • Give the URL as rtp://@:5004

A1 and B1 are the sending nodes (sending the video stream) In these terminals, run vlc-wrapper. In each opened VLC window do

  • Click Media → Stream

  • Add the video (from /usr/local/src/videos)

  • Click Stream

  • Click Next

  • Change the destination RTP / MPEG Transport Stream and click Add

  • Change the address and type to 10.0.0.2 in A1 and to 10.0.0.4 in B1

  • Turn off Active Transcoding (this is important to minimize CPU load)

  • Click Next

  • Click Stream

The video should be streaming across the network from A1 to A2 and from B1 to B2. If the video streams a slow or interrupted the CPU load is too high. In these cases either try a better machine or use a different (lower quality) video stream.

Take out L09 and let the Policy do it’s Magic

Now it is time to take out the link L09. This will be picked up by the Link Monitor, which sends a new event (L09 DOWN) to the policy. The policy then will calculate which customer should be impeded (throttled). This will continue, until SLAs are violated, then a priority calculation will kick in (Customer A is prioritized in the setup).

To initiate this, simply type link s5 s6 down in the Mininet console followed by exit.

If you have the video streams running, you will see one or the other struggeling, depending on the policy decision.

Reset the Demo

If you want to reset the demo, simple stop (in this order) the following process

  • Link Monitor

  • APEX

  • Mininet

  • Floodlight

Then restart them in this order

  • Floodlight

  • Mininet

  • APEX

  • Link Monitor

Monitor the Demo

Floodlight and APEX provide REST interfaces for monitoring.

  • Floodlight: see Floodlight Docs for details on how to access the monitoring. In a standard installation as we use here, pointing browser to the URL http://localhost:8080/ui/pages/index.html should work on the same host

  • APEX please see the APEX documentation for Monitoring Client or Full Client for details on how to monitor APEX.

VPN SLA Policy

The VPN SLA policy is designed as a MEDA policy. The first state (M = Match) takes the trigger event (a link up or down) and checks if this is a change to the known topology. The second state (E = Establish) takes all available information (trigger event, local context) and defines what situation we have. The third state (D = Decide) takes the situation and selects which algorithm is best to process it. This state can select between none (nothing to do), solved (a problem is solved now), sla (compare the current customer SLA situation and select one to impede), and priority (impede non-priority customers). The fourth and final state (A = Act) selects the right action for the taken decision and creates the response event sent to the orchestrator.

We have added three more policies to set the local context: one for adding nodes, one for adding edges (links), and one for adding customers. These policies do not realize any action, they are only here for updating the local context. This mechanism is the fasted way to update local context, and it is independent of any context plugin.

The policy uses data defined in Avro, so we have a number of Avro schema definitions.

Context Schemas

The context schemas are for the local context. We model edges and nodes for the topology, customers, and problems with all information on detected problems.

Trigger Schemas

The trigger event provides a status as UP or DOWN. To avoid tests for these strings in the logic, we defined an Avro schema for an enumeration (AVRO Schema Link Status). This does not impact the trigger system (it can still send the strings), but makes the task logic simpler.

Context Logic Nodes

The node context logic simply takes the trigger event (for context) and creates a new node in the local context topology (Logic Node Context).

Context Logic Edges

The edge context logic simply takes the trigger event (for context) and creates a new edge in the local context topology (Logic Edge Context).

Context Logic Customer

The customer context logic simply takes the trigger event (for context) and creates a new customer in the local context topology (Logic Customer Context).

Logic: Match

This is the logic for the match state. It is kept very simple. Beside taking the trigger event, it also creates a timestamp. This timestamp is later used for SLA and downtime calculations as well as for some performance information of the policy . Sample Logic Policy Match State

Logic: Policy Establish State

This is the logic for the establish state. It is the most complicated logic, since establishing a situation for a decision is the most important part of any policy. First, the policy describes what we find (the switch block), in terms of 8 normal situations and 1 extreme error case.

If required, it creates local context information for the problem (if it is new) or updates it (if the problem still exists). It also calculates customer SLA downtime and checks for any SLA violations. Finally, it creates a situation object. Sample Logic Policy Establish State

Logic: Policy Decide State

The decide state can select between different algorithms depending on the situation. So it needs a Task Selection Logic (TSL). This TSL select a task in the current policy execution (i.e. potentially a different one per execution). Sample JS Logic Policy Decide State - TSL

The actual task logic are then none, solved, sla, and priority. Sample task logic are as given below :

Logic: Policy Act State

This is the logic for the act state. It is simply selecting an action, and creating the repsonse event for the orchestrator (the output of the policy). Sample Logic Policy Act State

CLI Spec

Complete Policy Definition

The complete policy definition is realized using the APEX CLI Editor. The script below shows the actual policy specification. All logic and schemas are included (as macro file). Sample APEX VPN SLA Policy Specification

Context Events Nodes

The following events create all nodes of the topology.

Context Events Edges

The following events create all edges of the topology.

Context Events Customers

The following events create all customers of the topology.

Trigger Examples

The following events are examples for trigger events

Mininet Topology

The topology is realized using Mininet. This script is used to establish the topology and to realize network configurations. Sample Mininet Topology

2.3.0-SNAPSHOT Last updated 2020-04-03 16:04:24 IST