Number of CPE x sector

[starcomm2]

This is a question that can be answered in several ways, but I tought I'd ask.

How many customers I can expect to serve on each sector? I mean how many CPE could connect to a single sector (driven by atheros 2.4G) without them disconnecting or having issues. I don't take bandwidth into account (suppose there's plenty to share).

Is 50-100 on each sector a reasonable number?

Thank you.

[vjencic]

There are so many factors which may have great influence on No. of CPE's per sector (interferences, bandwidth ,polarization,platform -WRAP,WAR or PC,cloaking). My experience is up to 55-60.

[starcomm2]

I'm used to work with wrap boards (for firewalling purposes).

Do you think I should use WAR instead for an AP? How is stability and CPU power?

[gunther_01]

Start Reading the forum.

WRAPS are End of Life around here for new installs Go with WAR and don't look back

[tog]

I only do 30-60 clients per sector when selling bandwidth levels of around 1.5mbit - 3mbit

[go.fast]

Matt Larsen should chime in here, he has over 100 subs on a single ap.

He can give you the specifics of what he is doing.

[mlarsen]

I have 100+ subs on a 5ghz atheros based AP.

However, it is a Tranzeo AP that has 100+ customers on it, not StarOS. The most 5ghz customers that I have put on a StarOS AP is about 25. I'm using WRAP boards, and I'm guessing that the CPU will start having issues over 50 or so customers.

Matt Larsen
vistabeam.com

[Beebe]

I have 32 customers on a staros v3 WRAP board. It's running well. I think it's the AP on my network with the most subscribers, and it runs better than most other APs on the network since it's all StarOS clients on it.

I often see posts comparing number of subscribers to amount of CPU Horsepower. In other words if you need to add more customers, simply use a faster AP. However, I don't think it takes much CPU power to be able to max out 802.11a/b/g as far as number of clients go. I figure if you've got enough CPU to max out the throughput on a point to point link, then you've pretty much got enough to max out the number of clients that's possible. Of course if you add bandwidth shaping, layer7, firewalling, DHCP, routing protocols etc then that adds to the requirements.

I would think a WAR2 board would be able to fit as many clients as any other 802.11a/b/g solution.

[starcomm2]

WRAPS are End of Life around here for new installs Go with WAR and don't look backYes I know they're EOL and I already have access to the new design of the Wrap2. It's a good design and I know the designer personally. There will be 2 new wrap boards and one will be X86 compatible hence very easy to port Star-Os to. That's why I can't decide.

[lonnie]

What would the CPU have to do with the number of associations? The radio pushes the same peak data whether you have 1 user or 250 (which is our limit).

I have 100+ subs on a 5ghz atheros based AP.

However, it is a Tranzeo AP that has 100+ customers on it, not StarOS. The most 5ghz customers that I have put on a StarOS AP is about 25. I'm using WRAP boards, and I'm guessing that the CPU will start having issues over 50 or so customers.

Matt Larsen
vistabeam.com

[cephlon]

I had a Pentium M server running V2 with about 70 customers on it. It also did bandwidth shaping, firewall and handled 2 backhaul links. Last weekend, I moved the AP to a WAR4 and put a second compex card in with cloaking. I moved most of my V3 to the cloaked card (about 20). Moved the backhaul links to a WAR2 and put in one edge router that does all the shaping, firewall and NAT.

The WAR4 is handling 70 customers with no problem. Theres 50 on one sector and 20 on the other. CPU is around 2-5%.

I am also wondering how many more I can put on this one WAR4. Did someone say 250??

[therealboss]

Lonnie say the limit is 250, but I think it depends on what other 2.4 or 5.8 is in the area. I have one AP with 50+ subs on it but I think 40 or so is a good number for me.

[starcomm2]

Lonnie say the limit is 250, but I think it depends on what other 2.4 or 5.8 is in the area. I have one AP with 50+ subs on it but I think 40 or so is a good number for me.Woudn't 250 have hard time fighting for air time hence the network would collapse?

I agree that the CPU can pump more bandwidth on the radio card than the card can support, but I am wondering of there're hidden costs with many clients connecting. Bandwidth throttling priority queues (especially L7) are another issue to keep in mind.

[lonnie]

Of course 250 would pose a problem and I doubt you would have enough bandwidth, but, just the same, that is the design limitation.

L7 should NOT be done at the edge, unless it is right at the edge for a only a few clients. The CPU power required to examine each packet for the Layer 7 is more than I recommend you do on an AP or client radio. That is a job for the POP where you can throw a hefty machine at the task.

Woudn't 250 have hard time fighting for air time hence the network would collapse?

I agree that the CPU can pump more bandwidth on the radio card than the card can support, but I am wondering of there're hidden costs with many clients connecting. Bandwidth throttling priority queues (especially L7) are another issue to keep in mind.

[cephlon]

The edge router I meant is a Pentium M with 1GB of RAM. It serves all my customers, runs L7 and shaping, and hardly ever rises above 10% CPU. I guess my edge is my POP.

[starcomm2]

L7 should NOT be done at the edge, unless it is right at the edge for a only a few clients. The CPU power required to examine each packet for the Layer 7 is more than I recommend you do on an AP or client radio. That is a job for the POP where you can throw a hefty machine at the task.I would like to do preliminary shaping based on a token (filter-id) returned by the radius server. Can StarOs V3 do this while authorizing customers via PPPoE or PPTP (I could not find references in the manual)?

Also is it possible to share bandwidth un-evenly among different service levels (i.e. when the pipe is full give more bandwidth to higher classes instead of sharing evenly between high and low paying customers)?

[lonnie]

The PPPoE Server setup has a default bandwidth and from radius.

Bandwidth control is a simple maximum rate setting.

I would like to do preliminary shaping based on a token (filter-id) returned by the radius server. Can StarOs V3 do this while authorizing customers via PPPoE or PPTP (I could not find references in the manual)?

Also is it possible to share bandwidth un-evenly among different service levels (i.e. when the pipe is full give more bandwidth to higher classes instead of sharing evenly between high and low paying customers)?

[alaaghazi]

Hi Lonni,

Can you please give us practical figures from practical experience about the number of concerrent users on each AP?

Bye
Ala'a

[knolan]

We have 62 customers on a Wrap board with no problems

The Radio's are CM9's and we are running 5GHz
The CPU flickers between 3% and upto 20%

We have 30 customers on 1 sector & 32 customers on the other sector.

90% of these customers are on our smallest package 512k download, 128k upload.


The backhaul to the internet is via a WAR 4 connected to the Wrap via Ethernet.

The AP is running V3 1.1.13 and the only services that is running is OLSR

We are looking at adding another 40 or so customers to the AP, about 20 on each sector.


Regards,
Keith

[lonnie]

Practical is such a difficult term. There is a difference between connecting and providing service. Everything depends on several factors.

First is the speed which you allocate to each user. Then it depends on how that user makes use of that bandwidth.

If the user is doing "normal" type Internet activity, such as browsing or email then their use is bursty and you will be able to provide a high level of service to a large number of users. I would estimate that you could satisfy between 50 and 70 users with a 1 mbps allocation and they would be quite happy.

If you did not have limits and several of those users decided to run some file sharing software then they will consume all of the bandwidth you have available, and the number of users getting good service drops to between 2 and 5.

My recommendation is to use cbq to limit bandwidth right at the client CPE to some decent number such as 1 mbps. This is more than adequate for normal Internet tasks. If a user does consume everything, if you use cbq then they are only consuming their own 1 mbps and everyone else will not even notice. The standard AP can support 20 mbps or more so those few users doing 1 mbps continually will not affect anyone else.

If you wish to provide something like 4 mbps so they can do video on demand and file sharing I expect you'll have to limit each AP to about 10 customers, but even then you'll notice that not all 10 can consume their 4 mbps and you will have a problem if they try to do it at the same time.

[cephlon]

Only 10 customers per AP if you want to allow streaming Video? Are you talking Youtube stuff or HI-Def video? Alot of my customers are begging for faster speeds so they can watch youtube and other streaming video sources. But 10 per AP, I would run out of space on the tower very fast.

So I have a follow up question.
Lets say you only have space for a couple sectors on your tower. Can you use a signal splitter so that you have two cards on one sector? I mean, I know you can, but will this cause any problems? Does anyone do this?

[lonnie]

Youtube has low requirements. I am talking that the coming Industry norm will be 4 mbps and almost continual. If things are bursty then you can over sell, but if it is continual then you must prepare for lots more.

[starcomm2]

So I have a follow up question.
Lets say you only have space for a couple sectors on your tower. Can you use a signal splitter so that you have two cards on one sector? I mean, I know you can, but will this cause any problems? Does anyone do this?
You have a splitter insertion loss of about 2-3db. Other than this if you use different frequencies there should be no problems. However, try this before going to the tower. It's possible that the energy from one card is "sensed" in the other as a transmission in the air hence making it wait for "free time". Perhaps you'll need a couple of cavity filters between the splitter and the wifi card in the AP to reduce this effect (and they run from $150 - $200 US at least).

I've never done anything like this... I hope somebody that has will correct me if I'm wrong.

[cephlon]

Youtube has low requirements. I am talking that the coming Industry norm will be 4 mbps and almost continual. If things are bursty then you can over sell, but if it is continual then you must prepare for lots more.

The industry norm for consumer grade internet connection?

How do you plan for that?

[lonnie]

You actually PLAN for it. You do not install units based on price but will choose units that can provide the network design and control you will need in a year or two.

I used to pity the guys who kept installing CB3 and such but now I just figure they are creating a market for someone else to build it right. It is good advance advertising and a well designed network will easily displace the guys who do it cheap.

[cephlon]

I understand the hardware/software aspect of it, which is why I plan to slowing change out all my old units for WarBoards.

I guess I meant how do you plan for that with limited towers and tower space? If your tower can see over 500 potential customers, but you can only fit 10-30 on an AP, the tower wouldn't hold the amount of sectors you would need to service that area.

[lonnie]

You sectorize and you add more microcells. How many years have I been saying that? How hard is it to find a customer in a spot that will see another 30 customers? You distribute out the 2 port WARS boards on 5 GHz link and 2.4 GHz AP to a 15 dB omni. Using X2 cloaking you can easily repeat this as many times as you need.

Stop thinking that you must have a single tower with LOTS of power. That worked for the early guys but today it is required to have a bit better. Cloaking makes it possible and the multiple radio units are just what is required.

Personally I would put up a 4 port Metro at every possible repeat point and be able to extend using 900 MHz, 2.4 GHz or 5 GHz. The radios can be right there and you simply attach and antenna and aim it. Do a bit of tweaking in the config and now you connect to tanother micro cell from this one.

I guess people have not understood WHY were were so adamant about the 4 port boards. Some of you guys thought we were just doing it to get more radios than the competition. It is true we did have more usable radios than they did, but we went that route because our network design demands 4 radios.

[cephlon]

There are 2 issues with that setup that come to mind.

The first is power. What if you want to supply solar or battery backup to your network, so that when the power company goes down, you don't. I guess its not an issue if you live in a place where it never happens, but out here, if tree branch falls on the powerline, you could without power for 6 hours. Providing Battery/solar backup at each house is not as easy as managing it at a central site.

Don't you still run into an issue with bandwidth? If you set up these microcells, some may be 5 hops from the NOC. If you have 30 customers on each of those 5 hops, thats 4mps x 30 x 5. Seems like you would still run into a bandwidth problem with 4m constant.

[skyclimber]

Bandwidth is not a big issue. Not all users will use all their allocated bandwidth simultaneously.

I share 2x cable modem 10Mb/s . I have 3 hops of WAR2 with 100 customers: 256k/s, 1.5Mb/s, 3Mb/s, 6Mb/s and nobody is complaining about slow speed.