As part of a broader organisational restructure, data networking research at Swinburne University of Technology has moved from the Centre for Advanced Internet Architecture (CAIA) to the Internet For Things (I4T) Research Lab.

Although CAIA no longer exists, this website reflects CAIA's activities and outputs between March 2002 and February 2017, and is being maintained as a service to the broader data networking research community.


DIFFUSE for OpenWRT is a version of DIFFUSE that works on embedded devices such as home internet gateways. Our current prototype is based on the DIFFUSE Linux 0.4 distribution running on the Attitude Adjustment (r29537) version of OpenWRT (an embedded Linux operating system). DIFFUSE for OpenWRT allows you to enable automatic and dynamic QoS for your home network based on Machine Learning classification. Get the files here.


OpenWRT is a Linux distribution for embedded devices. It is commonly used to replace the factory firmware found on residential Internet routers. By replacing the factory firmware, users gain access to additional configuration options and are able to extend the functionality of the router by installing new software. The OpenWRT SDK was used to build DIFFUSE binary packages and DIFFUSE-OpenWRT firmware for the TP-Link WR-1043ND.

Although we have build and tested DIFFUSE only for the WR-1043ND, OpenWRT supports many home routers and it should be possible to build DIFFUSE for your platform using our build image.

The TP-Link 1043ND Home Router

DIFFUSE for OpenWRT was initially built for and tested on the TP-Link TL-WR1043ND Gigabit Ethernet residential router. It features one WAN port, four LAN ports and support for 802.11b/g/n wireless. It is based on the Qualcomm Atheros AR9132 SoC (MIPS) running at 400Mhz. There is 8MB of flash storage and 32MB of system RAM. DIFFUSE binary packages and firmware images for this router can be found in the downloads section. The OpenWRT page about the WR-1043ND can be found here.


Many home networks contain a mixture of traffic types (online games, video streaming, web browsing) as well as a mixture of access devices (PCs, Tablets, Mobile Phones). All these devices must share a common Internet connection. For access links such as ADSL and ADSL2+, the upload bandwidth of the link is often no more than 256Kbit/sec-1.5MBit/sec. Once the uplink link becomes congested (for instance due to uploading a video to Youtube, or putting a file into cloud storage), the performance of delay sensitive applications (online games, VoIP) can degrade significatly due to increased latency and jitter. A home router with DIFFUSE enabled can identify and prioritise delay-sensitive traffic dynamically.

See the DIFFUSE introduction for more information about the benefits of QoS and the DIFFUSE approach.

Traffic Prioritisation

A simple use case (explained further in CAIA Tech Report 120412A) can demonstrate the benefits of using DIFFUSE for QoS. In this example we configure the router to automatically classify traffic flows and then provide prioritisation for 'high-priority' flows. In the DIFFUSE architecture, this configuration can be described as a Classifier Node plus Action Node (CN+AN).  The access link has an upstream bandwidth of 1Mbps.

In our scenario one user is playing the online game Return to Castle Wolfenstein: Enemy Territory (ET), while another user is performing a bulk TCP upload. The ET traffic represents a real-time delay-sensitive application, while the TCP transfer represents a delay-tolerant flow (TCP is commonly used for delay-tolerant applications such as P2P filesharing). We measured the one-way-delay (OWD) of the game traffic flow (client to server) with and without DIFFUSE to determine the effect of the upload. We also measured the overall packet loss.

Figure 1 shows the cdf of the measured OWD. Without additional traffic the ET flow experiences very little delay. However introducing a single TCP upload increases the median OWD of the ET game flow to 351ms, and causes a 7% packet loss. A high OWD such as this would make the game uplayable. With DIFFUSE enabled the median OWD is reduced to 31ms, without packet loss. This is a significant improvement and would allow a user to play Enemy Territory while the file upload is in progress.

cdf of one
                            way delay from client to server
Figure 1: CDF of measured one-way delay for the online game flow.

Figure 2 shows the measured OWD for the ET flow against time when sharing the link with additional traffic, with and without DIFFUSE enabled. Without DIFFUSE the game flow packets are queued and delayed behind the TCP flow as they exit the router, seen here in the cyclic pattern of measured OWD values (ranging from 200ms to 500ms) as the upstream queue fills and empties. TCP will increase its throughput (congestion window size) until the queue is filled and packets are dropped. Once packets are dropped TCP will back-off, reducing the congestion window. With DIFFUSE enabled not only is the OWD reduced, the cyclic behaviour has been smoothed, meaning there is also less jitter experienced by the game flow.

                            way delay over time for game client to

Figure 2: Game flow one-way delay over time with TCP cross traffic. With and without DIFFUSE enabled


If you have a TP-Link WR-1043ND:

  • The easiest way to get started is to get the firmware image and flash your router with this new image.
  • If you are already running OpenWRT Attitude Adjustment on your WR-1043ND, get the packages.

If you don't have a TP-Link WR-1043ND:

  • Download one of the VirtualBox images and use the OpenWRT build tools to create an image for your router (everything you need to do this is included).
  • If you already have an OpenWRT build environment then use the DIFFUSE source code.

DIFFUSE requires a classification model to run. We have provided a model that will allow you to prioritise First Person Shooter games here.


  • The CAIA Tech Report 120412A  is a recommended starting reference. It details the memory usage, CPU performance and network throughput that can be expected when running DIFFUSE on the TP-Link WR-1043ND. It also expands on the use case described above.
  • HowTos. Guides on building/installing OpenWRT and DIFFUSE. The instructions are specifically for the TP-Link WR-1034ND but should work for other routers. Also contains some shell scripts for automatically setting up DIFFUSE and a sample classifier model. Direct links to the guides are also available below:
  • Startup script. A basic shell scripts that will load a classification model and configure DIFFUSE (the classification model is packaged with the startup script).

Firmware Image and Binary Packages for the TP-Link WR-1043ND

  • Firmware Image for the TP-Link WR-1043ND. Includes everything you need to get DIFFUSE running. Installation instructions can be found in the Scripts and HowTos.
  • Binary Packages for the TP-Link WR-1043ND. Tested with OpenWRT Attitude Adjustment (r29537). Includes all DIFFUSE components. These are opkg files that can be installed using OpenWRTs package manager.

Sources for DIFFUSE components

  • Source Files. Use these to build DIFFUSE packages for your router or existing OpenWRT installation. Building the source files requires the OpenWRT build environment. If you do not already have the build environment set up, get one of the Virtual Box VM images below (the source files are included with these VMs). You will need a version of OpenWRT that supports libsctp (we tested with OpenWRT Attitude Adjustment).

OpenWRT build environment Virtual Machines

  • VirtualBox Image (Light) (382MB). An Ubuntu Server 10.04 installation that is pre-configured with the OpenWRT build tools and all the DIFFUSE source code. A self-contained build environment that can be used to create new firmware.
  • VirtualBox Image (Build) (1.4GB). The same as the Light VM but has everything pre-compiled (CAIA subnet only).
  • Get VirtualBox here.

OpenWRT build environment under FreeBSD

  • Instructions on configuring FreeBSD as a build environment for OpenWRT.

Classification Models

Project Members

This project has been made possible in part by a gift from The Cisco University Research Program Fund, a corporate advised fund of Silicon Valley Community Foundation.


Last Updated: Thursday 21-Mar-2013 14:37:52 AEDT | Maintained by: Nigel Williams ( | Authorised by: Grenville Armitage (