Preseem co-founder and Chief Product Officer Dan Siemon recently put together a presentation to help ISPs learn more about capacity planning for speed tests in their networks.
You can watch the full video here or read below for a recap of what’s included. Also, don’t miss out on our free resources on this subject—make sure to download our eBook on Capacity Planning and the RF Environment, and our annual Fixed Wireless Network Report.
During the video and slideshow, Dan uses a very simple model with three main components: capacity, headroom, and an access point with six subscribers at a typical bandwidth. These examples are used throughout to examine and explain typical subscriber usage, access point capacity in a multi-point wireless network, capacity headroom, effective AP link rates, and more.
What is Typical Subscriber Usage?
Typical subscriber usage is the amount of bandwidth that a subscriber is going to use on your network during peak time. This is something that we report on in our annual Fixed Wireless Network Report.
In the most recent edition of the report, the median used download speed by plan rate across all of our customers was at around 6 Mbps, a 17% increase year-over-year. You can count on bandwidth usage growing each year, whether you want it to or not 🙂 It’ll grow even faster than this if you’re rolling out fiber, Tarana, or larger plans.
What’s interesting, as shown in the image above, is that the bandwidth consumed by users doesn’t grow nearly as fast as the plan speed. For example, if you sell a 50 Mbps plan, customers are not going to use double what they would in a 25 Mbps plan.
In fact, it’s much flatter than that. The chart above shows the slow growth in usage as the plan speed increases. For plans under 100 Mbps, for example, customers will actually only use about 5 Mbps on average. Even in plans of 250 Mbps and beyond, the upward slope is a lot less steep than you might expect.
Access Point Capacity
Understanding capacity is much more complicated in a multi-point wireless environment than in a fiber network or a backhaul link with good signals and constant modulation.
The data that Preseem collects each day from our hundreds of customers allows us to measure the real-world performance of access points from multiple vendors. As you might expect, this differs quite a bit from the idealistic case that’s published on vendor spec sheets.
As shown in our free Fixed Wireless Network Report, the 95th percentile of an AP’s performance represents a reasonable estimate for the peak performance of that model. This means that 95% of APs of a given model achieve less throughput than this during peak hours. For example, as shown above, 95% of the thousands of ePMP 2000 models that we see across all our customers deliver less than 60 Mbps during the busiest 10% of the day.
Headroom
Headroom is the gap between subscriber usage and capacity. As mentioned earlier, typical subscriber usage does not increase exponentially with plan size. And that’s a good thing because it means you can sell larger plans, e.g. you can double the plan without doubling the capacity on your network.
For speed tests, however, it’s a little different, as the plan rate really drives what the customer expects to see on a test. This in turn drives the headroom calculations. So, if you want to deliver a 20 Mbps speed test on your network, you’re going to need 20 Mbps of free capacity on that AP. If not, they either won’t hit the plan rate or will have a negative impact on all the other customers on that plan rate. So if a speed test kicks off and takes 80% of the bandwidth on an AP, then everyone else’s experience on the AP will suffer.
The slide above shows that with six customers on the AP at an average of 5 Mbps usage, you’ll need 30 Mbps headroom to get that 1:1 ratio of headroom to usage. As a result, a plan of around 35 Mbps is the max plan you can sell on this AP to be able to “guarantee” that a subscriber can achieve their rate on a speed test without damaging the experience of the other customers on that access point.
Now, our example above has nice flat lines but that’s not typically going to be the case on a multi-point network because of something called the Effective AP Link Rate.
Effective AP Link Rate
In Preseem, we have a model of this that looks at who’s transmitting, at what modulations, at what times, etc. That all gets combined to create a model of the Effective AP Link Rate at a given time. The slide below shows the rate for the same AP over different time periods—a roughly two-week period on the left and a two-day period on the right.
As you can see, the aggregate throughput changes between 40 and 80 Mbps over time. This happens because, during peak hours, it’s more likely that your low-modulation subscribers are active. As a result, your aggregate effective AP throughput is going to go down during these times.
As an aside, this is an example of the importance of good modulations on your access points. In the example below, we see an ePMP 2000 with very good RF conditions and 18 subscribers, all with high modulation. The effective link rate is up around 200 Mbps, which is obviously very high. However, because it’s more consistent, it’s much easier to capacity-plan this network and much easier to guarantee a good experience for customers.
How Would This Look on a Real Network?
Below is a more realistic picture of a network than our previous flat-line example. There’s a capacity line and headroom that’s shifting throughout the day, with the impact of low-modulation subscribers causing a dip as they come online in the evening. That’s also the time when more users come online and bandwidth demand increases.
On the graph, where you see the dip in capacity and the rise in demand shorten the amount of bandwidth available—that’s the most difficult time to hit speed tests. If a customer runs a speed test at that time of day and chews up some of that remaining headroom, that’s when it’s going to have the most impact on other subscribers.
This is one of the reasons why it’s so important to have a system managing the traffic that understands AP capacity, not just individual subscribers. Otherwise, you can’t modulate bandwidth and enforce fairness across subscribers.
So what does it actually mean to try and hit the plan rate at all times? Even the CAF Phase II funding for broadband buildouts has success conditions around hitting 80% of the bandwidth 80% of the time. In our example AP, this should generally be no problem—it’s only during the peak times when that’s going to be difficult.
Ultimately this becomes a business decision for ISPs, i.e. how much of the day do I need to be able to hit speed tests? This will determine your capacity planning so that you can consistently deliver a great experience to all of your subscribers.
Watch the full video below!
