When Xfinity announced their Xbox 360 application and said that the data usage would not be counted against our bandwidth caps, we became curious. We wanted to understand more about how the traffic entered our homes and how it might be treated in comparison to the rest of the traffic coming through the same plug on our cable modems. Here are the tests we ran along with the results.
First we wanted to see if there might be a separate channel configured for this Xfinity traffic on the cable modems that we received when we bought our Internet service from Comcast. We logged into the cable modem and looked at the configuration information that is made available on the modem’s web browser interface. The screen shot below shows the connection information from the cable modem. We are familiar with the DOCSIS standard that this cable modem uses and understand that channels are allocated in 6 Megahertz blocks and that the upstream and downstream channels are separate from each other. Looking at the cable modem configuration, we saw what appears to be just one of these 6MHz channels allocated for upstream and one allocated for downstream traffic.
This seemed strange to us because we understand that newer versions of DOCSIS allow multiple 6 Megahertz channels to be aggregated to create more bandwidth for the data channel. In search of additional channels we looked on other screens showing the cable modem configuration and found another screen that shows the total downstream bandwidth of the modem to be 42 Megabits per second. This amount of bandwidth is consistent with a single 6 Megahertz QAM256 channel.
We have also heard Comcast indicate that this last mile part of their network can get congested, so we were curious how this capacity was used by the Xfinity traffic when compared to other video on demand content.
In order to see how the capacity was being used, we set up a home network to be able to capture the packets that came into and left the house. To do this we put a router between the cable modem and the rest of the home network. We instructed the home router to mirror traffic from the port connected to the cable modem onto a laptop. On this laptop, we ran the Wireshark application to capture and analyze all traffic. While using this configuration, we ran the Xfinity application on the Xbox and then separately ran some MLB.com traffic. The packet captures below show that the Xfinity applications are marked with a diffserve marking of CS5 and that the MLB traffic was marked with CS1. We also ran tests for other video content including Hulu and Netflix, and saw the same CS1 markings on that traffic.
Before we move on to the testing for traffic that we conducted, we wanted to check to see if the Xfinity servers identified by the IP addresses above are a part of the Internet or if they are truly on a separate private network as Comcast’s statements appear to claim. To do this, we simply went to our work location and used that non-Comcast Internet connectivity to see if we could find those servers on the Internet. We executed a ping command to see if the servers were reachable and would respond. Sure enough they did. This implies that Comcast serves Xfinity traffic utilizing the Internet rather than a virtually or physically private network.
We then moved on to check to see what happens when the Internet link to our home is congested. We ran six types of tests and captured packets in the same way as described in the previous test. Each of the three tests below were run in situations where there was no other traffic from the home on the Internet connection and were also run with a fully utilized Internet connection.
- Xfinity traffic to the Xbox
- Netflix traffic to the Xbox
- Netflix traffic and Xfinity traffic simultaneously to separate Xboxes
To congest the Internet connection to the house, we used a software tool called Iperf.
We again used Wireshark to capture the packets and also used the information to calculate a throughput for each of the video streams. You can look at the data below and decide whether the Xfinity traffic gets preferential access to the home Internet connection.
Aggregating the data above shows the differential performance for delivery of Netflix and Xfinity traffic in congested and uncongested scenarios:
While there are a number of factors that can influence download performance for Internet traffic, it appears to us that Xfinity consistently gets good performance in both the congested and uncongested tests, while Netflix traffic is significantly impaired when the home connection is congested. These results seem to be consistent with the practice of prioritization.
Note: this post was co-authored by Andrew Dugan, SVP of Network Engineering & Architecture, and Nasser El-Aawar, Principal Network Architect.
