Predicting Broadband Usage on Networks

One of the hardest jobs these days is being a network engineer who is trying to design networks to accommodate future broadband usage. We’ve known for years that the amount of data used by households has been doubling every three years – but predicting broadband usage is never that simple.

Consider the recent news from OpenSource, a company that monitors usage on wireless networks. They report a significant shift in WiFi usage by cellular customers. Over the last year AT&T and Verizon have introduced ‘unlimited’ cellular plans and T-Mobile has pushed their own unlimited plans harder in response. While the AT&T and Verizon plans are not really unlimited and have caps a little larger than 20 GB per month, the introduction of the plans has changed the mindset of numerous users who no longer automatically seek WiFi networks.

In the last year the percentage of WiFi usage on the Verizon network fell from 54% to 51%; on AT&T from 52% to 49%, and on T-Mobile from 42% to 41%. Those might not sound like major shifts, but for the Verizon network it means that the cellular network saw an unexpected additional 6% growth in data volumes in one year over what the company might normally have expected. For a network engineer trying to make sure that all parts of the network are robust enough to handle the traffic this is a huge change and means that chokepoints in the network will appear a lot sooner than expected. In this case the change to unlimited plans is something that was cooked-up by marketing folks and it’s unlikely that the network engineers knew about it any sooner than anybody else.

I’ve seen the same thing happen with fiber networks. I have a client who built one of the first fiber-to-the-home networks and use BPON, the first generation of electronics. The network was delivering broadband speeds of between 25 Mbps and 60 Mbps, with most customers in the range of 40 Mbps.

Last year the company started upgrading nodes to the newer GPON technology, which upped the potential customer speeds on the network to 1 gigabit. The company introduced both a 100 Mbps product and a gigabit product, but very few customers immediately upgraded. The upgrade meant changing the electronics at the customer location, but also involved a big boost in the size of the data pipes between neighborhood nodes and the hub.

The company was shocked to see data usage in the nodes immediately spike upward between 25% and 40%. After all, they had not arbitrarily increased customer speeds across-the-board, but had just changed the technology in the background. For the most part customers had no idea they had been upgraded – so the spike can’t be contributed to a change in customer behavior like what happened to the cellular companies after introducing unlimited data plans.

However, I suspect that MUCH of the increased speeds still came from changed customer behavior. While customers were not notified that the network had been upgraded, I’m sure that many customers noticed the change. The biggest trend we see in household broadband demand over the last two years is the desire by households to utilize multiple big data streams at the same time. Before the upgrades households were likely restricting their usage by not allowing kids to game or do other large bandwidth activities while the household was video streaming or doing work. After the upgrade they probably found they no longer had to self-monitor and restrict usage.

In addition to this likely change in customer behavior the spikes in traffic also were likely due to correcting bottlenecks in the older fiber network that the company had never recognized or understood. I know that there is a general impression in the industry that fiber networks don’t see the same kind of bottlenecks that we expect in cable networks. In the case of this network, a speed test on any given customer generally showed a connection to the hub at the speeds that customers were purchasing – and so the network engineers assumed that everything was okay. There were a few complaints from customers that their speeds bogged down in the evenings, but such calls were sporadic and not widespread.

The company decided to make the upgrade because the old electronics were no longer supported by the vendor and they also wanted to offer faster speeds to increase revenues. They were shocked to find that the old network had been choking customer usage. This change really shook the engineers at the company and they feared that the broadband growth curve was going to now be at the faster rate. Luckily, within a few months each node settled back down to the historic growth rates. However, the company found itself instantly with network usage they hadn’t expected for at least another year, making them that much closer to the next upgrade.

It’s hard for a local network owner to predict the changes they are going to affect the network utilization. For example, they can’t predict that Netflix will start pushing 4K video. They can’t know that the local schools will start giving homework that involves watching a lot of videos at home. Even though we all understand the overall growth curve for broadband usage, it doesn’t grow in a straight line and there are periods of faster and slower growth along the curve. It’s enough to cause network engineers to go gray a little sooner than expected!

How Fast is Your Data Traffic Growing?

exponential-growth-graph-1I saw a quote recently from Jeff Finkelstein, the chief of the networks at Cox Communications, who said that the data demand on his networks was growing at 53% per year. I would hope that the managers of most large networks can cite their growth statistics.

There has been a metric in the industry that residential data usage has been doubling about every three years. This metric has roughly held true since the early dial-up days. Year after year people download more than they have the year before.

But it’s easy for us to lose sight that the bandwidth that people and businesses use for getting to the web is only a piece of the bandwidth usage on an ISP network. There are two other big uses of the networks that are growing faster than Web usage. We are seeing the first real growth in traffic from the Internet of Things, but the fastest growing contributor to web traffic right now is machine-to-machine (M2M) traffic.

M2M is when devices talk to each other. One example would be programs that automatically back up things to the cloud. PCs and cellphones now routinely send data to and from the cloud without any specific action by the user. The proliferation of storing data and using programs in the cloud has exploded the M2M usage.

The expected increase in M2M traffic is staggering as more and more things move into the cloud. In 2015 so far the average M2M traffic worldwide is about 50 terabytes per month. By 2018 that is expected to grow to over 900 terabytes per month. And while IoT traffic is relatively small right now, it is starting to grow rapidly as well.

Finkelstein says that the only way for his company to keep up with this fast growth is through the use of software defined networks (SDN). Cox uses SDN today to identify segments of traffic to route everything as efficiently as possible. He says that the company can isolate things like children’s traffic from parent’s traffic from business traffic and route each differently.

Just a few years ago there was a proliferation of peering arrangements established at companies like Cox. They created direct peering connections with companies like Google. But peering is getting more complicated and the goal for a large company is to peer with the major clouds – the Google cloud, the Amazon cloud, the Microsoft cloud etc. And that is where SDN comes into play to help route traffic as efficiently as possible to save on transport costs and to cut down on latency.

What this means for the small ISP is that the rate of growth of overall data traffic is accelerating. The 53% annual growth number would have seemed like an unbelievable number five years ago. For anybody not preparing properly the effects of that level of exponential growth can catch up to any network in a hurry. If Cox’s 53% growth is sustained they will have 5.5 times more traffic five years from now than they have today.

I don’t know how many network operators are planning ahead for that kind of growth. Certainly every network is seeing growth even if it’s not at quite the speed Cox is seeing. Traffic on rural networks is probably not growing quite as quickly as the Cox network, but it is still growing rapidly.

The major issue for most network owners will be to keep an eye on the various choke points in your network to understand where more data is going to cause you problems in the near future. Choke points can exist at many different places in the network from the backbone data pipes down to neighborhood nodes. And keeping all parts of your network  ahead of demand is going to require capital spending to upgrade electronics.