1. SDN API for Access Network Virtualization [1]
John Matthews, Craig Russell & Vijay Sivaraman (UNSW)
CSIRO Computational informatics
Residential internet data consumption continues to grow at 40% per annum while ISP revenues are growing at less than 4% per annum, mainly due to flat-rate pricing. Meanwhile, content providers, who monetize their video offerings via ad-based or subscription based models are seeing a direct impact on their revenue from reduced user quality of experience (QoE). This demo shows how ISPs could exploit the service quality dimension (in addition to bandwidth and download quota), using software defined networking (SDN) technology, to differentiate their offerings and tap into new revenue opportunities.
The demonstration
The demonstration network, shown in figure 1, is a greatly simplified emulation of content provider, ISP and home networks.
The home internet connection is emulated using 10Mbps Ethernet, which is considered in this demonstration to be the single bottleneck for service delivery into the home.
To show the hit and miss video quality available at most homes today we attempt to view a video on the big screen at the same time as a second 2.1Mbps (peak) video and a 20MByte file download.
Big screen video quality is poor due to it requiring more than it's TCP "fair share" of the home internet connection. Common workarounds, available today, are to re-schedule the file download to run during the night or to compromise on video quality by switching to a lower resolution on the big screen.
We show that it is possible for the user to achieve quality video on the big screen, under demonstration conditions, using our ISP API.
Home user control over QoE
The demonstration includes an example ISP toolbox web page with a single slider control to set alpha. Alpha is the maximum proportion of the home internet connection that can be made available for content provider slices.
Higher alpha results in better quality video on the big screen; values less than 96% result in a noticeable degradation to big screen quality in this demonstration.
Lower alpha results in faster file downloads; figure 3 shows the difference alpha makes to the average 20MByte file download time measured in this demonstration.
Table 1 shows results from a bigger 3 house experiment recently run at CSIRO. It can be seen that the mean and standard deviation of big screen video MOS and FTP stretch varied monotonically with alpha.
APP
AlphA=0.0
MEAN STD
ALPHA=0.8
MEAN STD
ALPHA=1.0
7.5Mbps video MOS
2.1Mbps video MOS
100MByte web load (s)
FTP stretch
Table 1: Effect of alpha on video, browsing and FTP
Figure 3: Average FTP download times
Content Provider control over QoE
The demonstration ISP uses OpenVSwitch, Linux HTB queues and the Ryu SDN controller.
The ISP API, implemented by a Ryu control script, exchanges JSON formatted text strings between the ISP and content providers.
Home video requests are handled by the video content provider, which knows the user's video quality expectations. This sends a bandwidth reservation request to the ISP API.
For example, to request a 7.5Mbps minimum bandwidth slice of the internet link to home device , the JSON format request would be: { hello: jukebox, type: minbw, nwsrc: /32, nwdst: /32, proto: 6, sprt: 8080, dprt: bw: 7500 } .
If the ISP accepts this bandwidth reservation request then the flow is assigned to a dedicated minimum bandwidth queue and the ISP may charge the content provider for delivery.
Similarly the FTP server, which knows the file size and the user's file transfer time expectations, sends a bulk transfer request to the ISP API. If the ISP accepts this bulk transfer request then the flow is given a dedicated queue, treated as "elastic", and the user may be charged at a lower rate.
Figure 2: Example home internet connection slices
Figure 1: The demonstration network
ISP API and Access Network Virtualization
Unlike previous QoS schemes, such as DiffServ, the use of SDN with an API provides much greater agility for the provisioning of services.
Our scheme requires no changes to the access network or home equipment.
Virtualization of the access network allows content providers to guarantee content quality without the complication of having to know about other traffic using the network.
The ISP could charge content providers for improved service delivery guarantees.
Homes retain control over the remaining best-effort slice which operates as at present.
References
[1] Vijay Sivaraman, Tim Moors, Hassan Habibi Gharakheili, Dennis Ong, John Matthews and Craig Russell, “Virtualizing the Access Network via Open APIs”, submitted to Conext 2013
John Matthews e w