In Episode 16, I mentioned that I had carried out an experiment comparing Verizon and AT&T networks. Below is an paraphrased excerpt from that experiment for interest.
In order to understand real world bandwidth allocation, I conducted an experiment to gather actual data rates from both Verizon and AT&T. In this experiment, I used two modern smart phones (Internet enabled, dual radio): a Motorola Droid running the Android 2.1 operating system on the Verizon network, and an Apple iPhone running the iPhone 3G operating system on the AT&T network.
I made use of an app (Internet applet) that was available both for the iPhone and for Android phones: Speedtest.Net. The Speedtest app simply finds a server in geographic proximity to the device, and then tests the available bandwidth in three ways. First, latency is tested by measuring, in milliseconds, how long a packet takes to travel to the server and back; in essence a ping test without the use of the echo protocol. Next, a file of significant size (a video, audio, or large text file) is downloaded to the phone. The app calculates the speed at which the download is taking place every second, then outputs the average download speed in kilobytes per second. Finally, the file is uploaded back to the server, and the upload speed calculated. Using these measurements a clear picture of the available network bandwidth is obtained.
My experiment consisted of 24 bandwidth tests–two an hour for 12 hours. The results, detailed in Figures 1-3, showed different available bandwidths across the two different networks, and highlighted how the different network architectures responded to heavy bandwidth allocation requests.
Analysis of download speeds, displayed in Figure 1, show that the Verizon network allows more consistent available bandwidth. However, the available bandwidth is much less, almost 50% that of the throughput that AT&T provides to the iPhone. There is something to be said for consistency; while AT&T allows blazing fast speeds skyrocketing to 3 MBps, it can bottom out just as often. Except during peak hours in the afternoon, Verizon maintained a fairly consistent distribution of bandwidth. These very small-scale results are consistent with the popular notion of AT&T vs. Verizon performance that can be found extensively in the media and in technology news (Laporte, 2010).
The upload speeds show another interesting effect of the backbone architectures of the two networks, as charted in Figure 2. The Verizon network surprisingly surpasses the available upload bandwidth of AT&T. This could be because Verizon makes no differentiation between upload and download channels–they are both the same consistent lower data rate. However, the data shows that the AT&T network attempts to maximize the download speeds (by far the biggest usage of data) by favoring download channels over the upload ones. AT&T uploads are so close to zero, that I would hazard that they might even use similar channels to the control channel (which is used for incoming call notification, push messaging, geolocation, and SMPP).
On average, the latency of both networks is very similar, as expected. The latency results of the experiment are detailed in Figure 3. The backbone networks are optimized in terms of routing, and latency has very little to do with bandwidth since ping is a very small packet size. However, AT&T spikes coincide with the drops in bandwidth in the download graph–meaning there was almost no available bandwidth during those tests. Again, the consistency of the Verizon network and the lower latency with spikes of inoperability for AT&T is consistent with the popular language concerning network architectures in the media (Vogelstein, 2010).
There are two types of backbone networks: a backbone network that uses lower cost base stations at a much greater geographic dispersion, and a backbone network that implements high throughput, more expensive base stations at the cost of lower coverage. I believe that my experiment results show that Verizon has implemented the lower cost base stations, trying to maximize coverage, while AT&T has implemented the higher cost, fewer base station approach. Although this idea has been made popular by advertising and technology news reporting, the data clearly indicates that the goal of the addition of relay stations is different for each type of network–Verizon is attempting to add additional throughput, and AT&T is attempting to minimize the latency spikes that plague its users.