How to know when an access point is full?

Hello,

I would like to know if there is something I can check to understand when an access point is full and you shouldn't add other SM.

I mean: you can put 120 users on a single sector, but it depends on the bandwidth you're offering and their quality, right?

I made some calculation using the Excel file in the support section and I had some numbers, but they're theoretical.

I currently limited my access point to 80 users (from my calculation I should go up that number), but is there a way to check if my access point is performing bad due to over-subscribe, so I know it's time adding one more sector in that direction?

Is there any suggestion about frequency on two access points in the same direction? I think the best is different frequency (you can't reuse in the same direction) and at least 5MHz spacing from other sector, right?


Thank you :)

If I remember well, up to 120 SM can connect simultanesly but it's  recommended to don't go over 40.

>giuseppe4 wrote: I mean: you can put 120 users on a single sector, but it
>depends on the bandwidth you're offering and their quality, right?

Hi. Well, here is my 2c opinion is this:

First of all, Cambium does a much better job, and a much more organized job of managing it's clients than anything else in it's class. The AP does the scheduling, and tells each client when it should listen and talk, and that eliminates 'hidden node' that eliminated 'near far', and that eliminates a ton of problems. Secondly, Cambium's scheduler is an 'Air Fairness' scheduler - so if you have a client who get's spun by the wind for example, and his signal drops from a -60 to a -77 because he's misaligned, Cambium again does a better job by still limiting him to his time slot. That client's performance will suffer, but he won't dominate the Access Point and squeeze everyone else out, which is what can happen on many other products. With other products, if that -77 client is trying to transmit, and if he's set to 1Mbit upload - well, he can still get it, even if that means he's using 40% of the AP's time and impacting everyone else.  All these things (and more) translates into being able to scale to many, many more clients than most products can - and to have more predictable performance as you do so. As I mentioned in another thread before, at the last Cambiumn ePMP Training course I was at, there were WISP's there from Alberta, and they had sectors with 60-80 clients on them, and they were very, very pleased with the performance.

That all begin said - YES, if you have all your clients installed at -75 and if you want to give everyone 20 Mbit down and 5 Mbit up, and if everyone streams 90% of their day at their full 20 Mbit - then no, you're not going to fit 80 people doing that all at the same time without something giving somewhere. If everyone is connecting at MCS 10 due to alignment or interference issues - then that will impact the amount of work that can get done per frame. But, if everyone is perfectly aligned, they are all connecting at a -55, they are all MCS15 all the time, and have packages where your typical peak load is within that the AP can do - then you should be good.

Also - the 'Air Fairness' algorithms do a pretty good job of 'fairly' sharing an AP when it's overcapacity too. What I mean by that is - with other products, if the AP is capable of 25 Mbit total capacity, and if you have 20 clients and each is set to a 3 Mbit plan, then bad (and unpredictable) things can happen when that 11 or 15 or 20 of those clients try to use the AP at the same time. The first 5 or 7 client might do fine, but as things approache over capacity, WiFi can start 'out competing' each other in a way. Actually, the clients start 'backing off' when they see contention - so they listen to see if the channel is clear, and if it's not, they back off a random amount of time, wait and then retest to see if the channel is clear again. If it isn't, they will back off for a longer random period, and retest again. That's an OK mechanism on your home router when you probably have 5 devices connected, and when your laptop can hear the tablet in the next room - but in an outdoor network, what that often results in that your full AP will be servicing 10 of it's clients OKishly, and when the 11th or 15th or 23rd client tries to get some air time, it can't get a word in edgewise. YES, there are mechanisms like RTS/CTS which do (kinda) help - but they also lower overall throughput too, so while they are solving one issue, they are creating another issue in the process.

With Cambium, the AP has scheduled all the SM's when they should talk, when they should listen - so a full AP with lots of traffic won't have SM's trying to 'butt into' the conversation. They will have been given a schedule of when their turn is, and your sector will listen for them at that instant. There still can be competitive interference, BUT self-interference or contention at that time slot will be minimized or eliminated in a well designed Cambium network.

And if the AP is simply over capacity, with more data being requested than it can really handle, then Cambium does a reasonably good job of sharing the air fairly - so if you have 40 clients on 5 Mbit plans, and if the sector has 100 Mbit that it can handle, and if all 40 clients are trying to download - then all 40 obviously can't get their 5 Mbit each. But, Cambium ePMP will do a pretty good job of 'fairly' giving everyone 2+ Mbit each of the available 100 Mbit capacity, instead of many other products, where some clients are getting 5 Mbit, and other clients are nearly dead in the water. And - low signal client or far distant clients are still a challenge, BUT they don't single handedly ruin the whole AP's performance. Thier less than ideal situations impact them mainly, and they don't get more than their fair share of air time.

>Is there any suggestion about frequency on two access points in the same direction? I think the best is
>different frequency (you can't reuse in the same direction) and at least 5MHz spacing from other sector, right?

Yes, GPS Syncing and TDMA and having matching Download/Upload Ratios and matching distances and so on - that's all about having matching TX/RX frame lengths and that's all about eliminating self-interference. You see, as a WISP, you deal with two main kinds of interference - competitive interference, and self-interference. Anyone who has done this for a while can tell you two things. A) self-interference is MUCH closer and louder, and B) self-interference is ultimately the only thing you can really hope to control.

Of course, there are ways to mitigate competitive interference by picking the correct frequencies and by using good isolationist, shielding, the correct antennas, etc - but at the end of the day, the competition, and the video surveillance system and the baby monitor are all interfering and you can't do too terribly much about it. But, they are also probably a -78 and they are not right in your beam and so on. However, self-interference can be a -35 and it can be 5 feet away and it IS something that you can 100% control if you design things correctly.

So, if your client is going to be able to hear two sectors (either adjacent sectors, or sectors broadcasting in the same direction) then you're going to want to have as much separation as you can. With GPS Syncing, that can be as little as 5Mhz 'guard band', but if you've got more, you certainly can use more spacing. Without GPS Syncing, the guidelines are to have 2x the guard band as the carrier band - so if you're using 20Mhz channels, you should have 40 Mhz of guard band. Again, that's not a horrible idea even if you are syncing everything properly.

12 Likes

Sure would be nice to graph the utilization %.

2 Likes

Hi ninedd, 

That's a lot more than 2 cents so really appreciate taking the time to knowledge transfer ePMP architecture to the community members.

This itself makes me want to push the developers even harder to give you your per SM real-time throughput monitoring on the GUI :) Not promising it yet though. 

Thanks,

Sriram

3 Likes

@jluthman wrote:

Sure would be nice to graph the utilization %.


Hi Josh, 

Yes, this is pending work too. Just a lot of things going on right now. So hopefully soon. 

Thanks,

Sriram

1 Like

ninedd, your answers are really helpful! I really appreciate the time you're spending for me in the forum (and for others). It's nice to see a great community helping newbies like me :)

It's all very clear: I can put on MikroTik (the vendor I'm switching from) 25-30 users if all MIMO and correctly pointed with 10/1, and they run not very bad until... self interferences, misalignment due to wind, competitive interference, and sooooo on!

So, expecting very close clients with very good signal in MCS14/15, I hope to reach higher numbers (60 would be a very good result, 70-80 excellent). Anyway, I think Cambium ePMP will help in self-interferences: I have a lot of AP because I can't add more than 25-30 users on one, so I had to make as much AP as possible, but I noticed this isn't a "future proof" design, so I decided to switch to Cambium ePMP. 

Anyway, a graph of air use from Cambium could really help, so I know when it's time to install a new sector, instead of having people complaining about their low speed :)

Siriam, hope you'll also push development team about burst in QoS :-D

2 Likes

@giuseppe4 wrote:

Siriam, hope you'll also push development team about burst in QoS :-D


Hi giuseppe4, 

Yes, I'm well aware of your request. I will try :)

Thanks for your support,

Sriram

1 Like

These first days with cambium community are great!
You’re listening your customers, and that’s very appreciated.

Thank you! :slight_smile:

This itself makes me want to push the developers even harder to give you your per SM real-time throughput monitoring on the GUI :) Not promising it yet though. 

I think the above is very important as it allows you to watch the throughput when someone calls tech with a speed complaint, we have a lot of clients who complain about speeds and its becuase they have something downloading in the background so there speedtest is showing inacurrate results of what there link is capable of.

So with a live throughput in the GUI you can see the traffic the unit is doing while they perform a speedtest etc, also you can watch the live thraffic while they isolate the problem and tell them as soon as the traffic stops.

So how atm do you tell what throughput the SM is doing?

Something like UBNT's aircontrol would be awesome, its one of the reasons I am not convinced a move is right yet. Least with UBNT I can monitor my whole network, perform mass sceduled FW updates etc and see live and past history data. 


@>Chris 1 wrote:

> 

>This itself makes me want to push the developers even harder to give you your
>per SM real-time throughput monitoring on the GUI :) Not promising it yet though. 

> 

>I think the above is very important as it allows you to watch the throughput when someone
>calls tech with a speed complaint, we have a lot of clients who complain about speeds and its
>becuase they have something downloading in the background so there speedtest is showing
>inacurrate results of what there link is capable of.

> 

>So with a live throughput in the GUI you can see the traffic the unit is doing while they perform
>a speedtest etc, also you can watch the live thraffic while they isolate the problem and tell
>them as soon as the traffic stops. 


Exactly Chris1 - that's exactly why we need it, and exactly why we're looking for it. We get that call numerous times per day, and right now (with our older Non-Cambium gear) we can see a fairly real time list of everyone on the AP, and not only their KBit totals, but also those totals dividen into KBits/second.  Each SM already has a fairly real-time (as little as 2 sec) display of the Total KBits through the connection, but that's not the same as having that translated into KBit per sec or KByte per sec or MBit per sec.  And seeing it on each SM is good, but that is also not quite  the same as seeing the display on the AP either - so that we can see what everyone is doing at the same time.  Fred calls (customer #6 of 40) wondering whyit's slow at the moment, well if the other 39 people are all transferring at 3 Mbit per second each... that's 100+ Mbit used up by those other 39 clients, and that's maybe why Fred is feeling that things are slower than normal.

Hello guys,

I set a graphic monitoring of some values from APs.

Green = UL Frame Used (there's a bug in ePMP 1000 and it's always near 100%)

Blue = DL Frame Used

Black = Number of users

I have now 6 users.

Each user have 10/1 profile and MCS14/MCS15.

As you can see, DL Frame Used reached 34.70% 1 hour ago (each point of the graph is 5 minutes from the last one).

In that moment, the total download traffic on the AP was about 20Mbps.

If 20Mbps = 34.70%, and there are only 6 users, does it mean I have to worry about total capacity?

100% would be about 60Mbps in download.


Am I right?
Thank you :)

PS.: How can I monitor real traffic per second via SNMP? I can't find a value for that!

I'm currently monitoring it from the router port.

Can we graph UL/DL frame used as a percentage now?  Is there an OID for this (percentage) in 2.6.1?

No, I made some math manually using PHP+SNMP and the graph is the result of that

1 Like

Assuming 60 mbps may, or may not be correct. A lot of things come into play when deciding it’s full. If in your performance tab you see “capacity drop” packets, you flooded over. Retransmission, errors nacks and packet size come into play when your talking about a sector filling up. Watching your frame utilization is the best thing, but it’s not graphable directly yet. You can record total available frames, and used frames. You’ll have to do math to see where you sit. Back to bandwidth, you may have one fill up at 55, the next may fill St 80. They unfortunately won’t land in the same spot normally. Unless of course, you’ve managed to get the exact same conditions twice, or everyones locked at mcs15 (wouldn’t that be great!?!)


@giuseppe4 wrote:

No, I made some math manually using PHP+SNMP and the graph is the result of that


Can you share in more detail about how you created that graph?

Sure,

you have to periodically execute something like that on your server (using cron for example), that should be able to reach the Cambium ePMP (so, VPN or an internal server):

$hostIp = '10.0.0.1';
$community = 'public';

$session = new SNMP(SNMP::VERSION_1, $hostIp, $community,3000000,5);
$session->valueretrieval = SNMP_VALUE_PLAIN;
$session->oid_output_format = SNMP_OID_OUTPUT_SUFFIX;

$dlFrameTotal = $session->get('.1.3.6.1.4.1.17713.21.2.1.52.0'); $ulFrameTotal = $session->get('.1.3.6.1.4.1.17713.21.2.1.51.0'); $ulFrameUsed = $session->get('.1.3.6.1.4.1.17713.21.2.1.53.0'); $dlFrameUsed = $session->get('.1.3.6.1.4.1.17713.21.2.1.54.0'); $dlFramePerc = number_format(($dlFrameUsed/$dlFrameTotal)*100,2); $ulFramePerc = number_format(($ulFrameUsed/$ulFrameTotal)*100,2);

//Here you should insert value in a table in a database

Above you have a PHP example, so you should have PHP >= 5.3.3 compiled with SNMP support + Apache/nginx on your server.

Then, you should graph the results you have in your db table (for example using MySQL as your DB) using some graph library, like jQuery RGraph or something similar.

And you should get something like this:

2 Likes

Great, this is actually a graph that a company has built into a package for us that is charing a ton of money for. Good to know its Semi easy to duplicate with the right equipment.

1 Like

You could create a VPN between your gateway and one server for example on DigitalOcean or something like that.

With a very low monthly expense you can track lots of data from your AP! :)