How many customers on an ap? wrong question

Several years ago, I did an article on How many customers can I fit on an AP? I figured with the introduction of MU-MIMO and other things, it was time for an update. Several concepts still apply, but we now have Multi-User MIMO, better filtering, and better technology. One of the biggest questions I hear is, “How many customers can I put on an Access point?”. In this article, I will explain some of the ways to answer this question. Some of this will be geared toward certain products but will be an overall way of answering the question.

Thinking in terms of how many customers you can put on an Access Point is flawed thinking. What you really should be thinking of is how much capacity do I have to sell on an AP. From this, you can apply a formula to know how many customers an Access Point can support with quantifiable data.

Firstly, some things to know.  This article applies to mainly point-to-multipoint radios.  Most of your multipoint radios you come across are half-duplex radios.  The radios receive or transmit, but not at the same time. The over the air rate vs. real throughput come into play as a result. More on this later. Before we get into everything we have to know what affects the customer data rates.  I will break this into two sections. Ideal environment and the real world.

The Ideal Environment
This mainly has to do with radio specs and such.  You have channel width, data rates, and signal to noise to worry about.

Channel width is the first thing to consider. The bigger the channel, the more bits you can flow. If we want to use an analogy, we could compare this to a road or a water pipe. The bigger the road, the more cars that can drive down that road at faster speeds. A larger water pipe can flow more water. As with anything, there are drawbacks. The larger the channel, the more susceptible you are to interference.

Data rates and modulation are the next factors.  The higher the data rate the more capacity the client radio has.  Data rates are influenced by the channel width, radio limitations, and environmental factors.  Think of data rates as the top speed of your client radios. Just like a car road conditions are a huge influencer.

Signal to noise is one of the most critical factors overlooked. I have included this in the ideal and real-world sections for a couple of essential reasons. In the ideal environment, radio manufacturers publish the signal to noise needed to achieve max modulation. Modulation should be looked at first when it comes to a radio not performing as well as it should. The first thing I always look at is what is the current signal to noise.  For example, a Cambium 450M (Medusa) access point states,in the Spec sheet, that in order to achieve an 8x modulation, which is 256QAM you have to have a signal to noise ratio of 32dB.  This chart means if your noise floor is a -80, you have to have a signal of *at least* -48.  In the real world, this isn’t always achievable. Physics can fickle that way. If you want to geek on what QAM is you can watch the following video

The real-world environment
As many of you know the real world can be totally different than the lab environment.  Let’s discuss some factors which can alter the modulation rates, which then affect your overall throughput on an AP.

RF Landscape of a link

The RF “landscape” is the most significant influencer. In other words, how noisy is the spectrum? How many other devices does your access point “hear”? I always use the crowded room analogy. If you have a couple of people in a room, it’s easy to hear them and more comfortable to talk faster (modulation rate). As more people enter the room, you have to find a corner with a smaller group to talk (change channels). As the room becomes even more crowded, you have to speak a little slower because those around you are noisy and a distraction. Your modulation rate has to lower to have an intelligent conversation.

Line of sight is the next major issue. If a customer has any obstruction between them and the AP, the modulation level to drop because it has to deal with the extra noise. This is simple physics. Not only does the signal get degraded if it has to pass through objects or even dense air, but it is also deflected. This deflection is referred to as multipath. Other factors that influence modulation are the quality of antennas, the quality of any cables between the antenna and the AP, environmental factors such as bodies of water, and many other items. these are beyond the scope of this article.

On to determining the total capacity of an AP

Let’s take a Cambium ePMP 3000 ap as an example. This is a 4X4 Multi-User MIMO radio.   What this means is it can transmit four streams to a user at once.  This increases the bandwidth to the client. So where does the multi-user part come in? Most clients are not able to take advantage of the Access Point’s (AP) full capacity so the AP talks to multiple clients at once because it has the capacity to do so.

So let’s run some numbers.  The published spec sheet of an ePMP 3000 radio is a total capacity of 1.2 Gbps.  This radio is a TDD system. This means you over the air rate is half of your actual throughput due to the half-duplex nature of the radio.  It can only send or receive at one time, not both.

Now that we know our radio will do approximately 600 megs of capacity minus some overhead we can factor in oversubscription.

Oversubscribing in the ISP world has been going on since the dial-up days. When managed properly, it is not a bad thing. The theory is that not every user is online at the same time doing the same things. Out of ten households doing things on the Internet at any given moment in time, you may have three or four streaming Netflix, two watching Youtube videos, three checking Instagram/Facebook/Twitter, and one just reading webpages. Let’s say each of them is paying for a 25 meg down by 5 meg up speed package. Out of those 10 accounts the Netflix streamers may be using 5 megs, the Youtube watchers may be using 3, and the rest are using a combined 5 meg. Out of 250 megs of sold capacity, those 10 accounts only use 31 megs at that point in time. Out of those users, only the streaming services are using that bandwidth the most. In an earlier article, I did a video on a Netflix stream at my house. As customer plans have more bandwidth available, they are grabbing data less frequently because they can grab bigger chunks at a time. This blog post illustrates this as well as this video

Here is where oversubscription becomes a moving target. Not every household is the same. Some may have two or three devices that stream at the same time.  Some may only have one.  Some may watch streaming services very little.

So how do you plan for oversubscription?
In today’s world of streaming a 3:1 oversubscription ratio is a pretty safe bet.  Depending on your customers you might be able to go 4:1, 5:1, or even more.  The faster your plans the less time the customer gets on and off the connection.

So let’s put it all together.
600 megs of AP capacity at a 1:1 ratio
1200 megs of AP capacity at a 2:1 ratio
1800 megs of AP capacity at a 3:1 ratio

For easy figuring, we will say we are selling 20 meg packages.
1:1 we can sell 30 20 meg packages
2:1 we can sell 60 20 meg packages

Will these numbers hold up in the real world? In most cases, they will not due to the real world conditions mentioned earlier in this article.  If you keep all of your customers at high MCS rates you should expect 70-80 percent capacity numbers in a real-world scenario.  Your mileage may vary. So let’s adjust our numbers.

70 percent of 600 megs is 420 megs
420 at 1:1
840 at 2:1
1260 at 3:1

Those same 20 meg packages
1:1 we can sell 21
2:1 we can sell 42
3:1 we can sell 63

Is the above formula absolute? It is just designed to give you an idea. The following link was published today. it shows 72 ePMP clients on a single AP. As I have stated the client connection isn’t the whole story.  Look at the throughput running through the AP to illustrate the formula is highly dependent on your customers and how they use the service. Remember when I talked about channel width and data rates? Pay attention to these in the video.

In conclusion think of how much capacity you have on an Access Point instead of just customer numbers.  The numbers can be impressive, as in the above video, but don’t tell the entire story.  Customer counts on an AP are nice to know and you can take the above formula to determine how many you can put on at what levels.

#packetsdownrange #epmp #rfelements #cambium




Water tower install with mounting frame

We recently headed up a job for a client of installing some RF elements horns, Cambium ePMP, and Baicells LTE for a client.  One of the gems of this job was the frame the client designed for the job.  We can’t take credit for this. We just think it’s cool. Some of these pictures were taken during construction, thus post clean-up.

The frame is truly an example of how WISPs are stepping up their installs to become more standardized and carrier-grade. It costs some money but is worth it in the end.


Quick tidbit on Cambium GPS sync source

Synchronization source– Synchronization mode supports two types of timing sources: the timing signal provided by a Cambium Network’s Cluster Management Module (CMM) or by the AP’s own on-board GPS receiver. Both these sources require GPS Synchronization capable AP hardware and will allow APs to receive a stable timing signal to synchronize their own Tx/Rx cycles.

There is a third option for the Synchronization source, namely “internal”. This “Internal” Synchronization source option is NOT an actual synchronized timing source and when selected, it just provides timing from an internal unsynchronized source. “Internal” is NOT to be used in GPS Synchronized deployments but could be used with single PMP AP installations. The “Internal” timing source is also used when the Flexible DL/UL ratio option is selected.

Cambium ePMP compatibility graphic

Please note the following for e3K compatibility. We do not recommend or support running a mix of 3.X AND 4.X on the same sector. All APs/SMs should either be running on 3.X release or 4.X release but not mixed for commercial deployment. Standard upgrade path should still be followed, 


Cambium 4.4.0-RC23 and some people losing their minds

For those of you not in the Beta program Cambium has broken tradition and widely announced a new RC for ePMP on some of the Facebook forums.  For those of you running ePMP, this is a significant upgrade for just an RC.  Some highlights for me in the changelogs.

Keep in mind 4.4 is a point upgrade.  For those of you who don’t know you have Major Versions (ie 4.x), Minor (4.4) and Patches (4.3.2 and hotfixes (   On top of these numberings, you have different builds Alpha, Beta, Release Candidate (RC), and others depending on the vendor. I expect an RC to have a higher build number by the time it is released to even Beta testers.  This means the internal team and their test networks have gone through some pretty solid feedback loops.  So on to why 4.4 is an important thing, especially to ePMP users. 4.4 is a minor release with significant improvements.

•QoS ePMP3000 now supports Quality of Service to support traffic prioritization. This is the first phase of QOS implementation supporting traffic prioritization at the individual SM/AP. Future additions will cover features such as station priority.

•ePMP3000 now supports maximum information rate further allowing the operator to manage traffic profiles for end customers.

•This release now supports trial configuration that allows a user to try a configuration change without applying the configuration.

•This release now supports cnMaestro Zero-Touch.

So let’s talk about each of these.
QOS.  This is where we can get into some debate. How much QOS should you be doing at the Subscriber module? How much of it should be handed off to an ISP managed router? I am under the philosophy an SM should be doing as little overhead of non-radio things as it can.  If you want to do QOS do a very basic QOS set, if at all.  If you do not do managed routers, then you almost have to do QOS at the radio.  Same goes for the traffic profiles. This is something you must decide for your self.  I have my philosophies and gladly share them. What works for me may not fit with your methods.

Trial Configuration. This has been a feature of UBNT that has been super awesome. You can apply your settings and if you lock yourself out you just have to wait until the settings are reverted.  I have not tried this yet, but it is an exciting thing.

Zero-Touch.  This has been a long time coming.  This reinforces the cnMaestro ecosystem this much further. At one point you will have enough automation setup where a new AP that comes online will have all of its settings pushed down to it without you ever having to log in to the device.

Bugs Addressed in 4.4.0-RC23
•FC ACG-5846 [FC] F300-25 could not connect to e2K in first attempt. Cause – INA RETRIES EXHAUSTED
•[F130] enable IPv6
•Downlink Frame Utilization always shows 0%
•Fragmentation improvements
•[MIR] MIR Profile Rules does not work

Known Issues in 4.4.0-RC23
•[FC] SMs are disconnected from AP with Reason: 48 (COMMUNICATION LOST) when traffic is transmitted
•PPPoE MTU size limits should depend from Ethernet MTU

What are some of my tips for getting the most out of your ePMP deployment (or any vendors deployment)?

If you are running 3k don’t run mixed mode in production just yet unless you are prepared to submit tickets and bug reports.  Also, 3k users run the latest beta firmware. Signup for the beta and READ the changelogs.  One of the things on a recent thread on the Facebook Cambium Users group was some users being vocal about a small number of things fixed.  If they read the changelog you would see a lot more was mentioned.

A good rule of thumb is your APs and SMs should be the same firmware version.  Not everyone in your whole network has to be the same.  You can have one tower or sector on version 4.4.0-RC23 and another sector on  The key is each SM attached to an AP should be the same version as the AP.  Once you start mixing weird results can happen.

Read the changelogs. In the current changelog for 4.4.0-RC23 I count 47 bug fixes and 14 known issues, some of the highlights I mentioned above. With a point upgrade like this, I would expect some major improvements and lots of them.  I always like to mention highlights in my posts like this because it helps those of you who are busy to pick out some significant things which may interest you. There may be some I don’t mention which are essential as well.  You should have someone on your team who reads the changelogs. Reading changelogs can save you much heartache, especially when it comes to known issues.  One of the things I like about the Cambium changelogs is they mention the tracking number.  If you have opened up a case and it is a verified bug you will receive a tracking number.  This tracking number allows you to keep the vendor accountable.  If you don’t see this in your other vendors, push them to do this.  The vendor may already be doing this internally.  Most software devs have ways to track their bugs internally. Push them to publish known and fixed issues, with tracking numbers in their changelogs.

Some of the issues in changelogs may never affect you and may be nitpicky things.  A good vendor will not “pad” their changelogs with fluff like “changed a spelling error here”. These are fairly insignificant and just add to the reason why some users ignore changelogs.

How has this helped me in the real world? Knowing what was fixed in a changelog allows me to decide if this is a software version which needs to be rolled out. Do the pros outweigh the cons? Is this a bleeding edge platform where I need to be running the latest in order to get the best benefit? You will find the new kid on the block is the one who gets the most development time.  The platform needs to reach a certain level of maturity as fast as it can.  If you have these in production or are testing then running the latest code is the way to go. As the platform becomes more mature, running the bleeding edge code isn’t as desirable.  You reach a point where you can wait longer between software upgrades.

I have always been of the philosophy where my devices are doing as little as they possibly can.  What do I mean by this? I have dedicated functions to devices in the network.  My Access Points are not doing firewalling, QoS, and other things.   If they are it is at a very basic level. I want each device to excel at its function. This leaves more horsepower to do its core function(s).  It also lessens bugs and conflicts between software processes.  This is where many of the bugs show up.  If your device is doing 4 things, then you have times the problem than if it’s doing just one thing.  However, this is the real world and you have to stretch your budget as far as you can.  Just be aware of adding more functions to a device can cause compatibility issues.  Keep It Simple Stupid (KISS). Network design and philosophy play a huge role in how a device behaves.

By combining all of the above you can work out an internal philosophy which allows you to tailor it to your business needs.  If you are running the latest hardware platform in production for whatever reason then have it do dedicated functions until the software is mature.  This will lessen the impact of bugs.  As the product matures you can turn on more functions.  Turns the dials and knobs as some would say.  If you are just testing it then turn it all on and try and make it break.  But submit tickets to the vendor.  This may seem like you are doing “their job” but this is part of the investment in the platform.

While this article started out being about Cambium, you can apply the philosophies to any vendor you want to put in your network.