Flow of Throughput...

For the past 6 months, my head has been crammed with information that I’ve never or rarely dealt with before, so in light of this please forgive these seemingly elementary questions:

I think I’m not fully understanding the flow of throughput within this Canopy system, so PLEASE correct me if I’m wrong. The way I see it is this:

Bandwidth flows to the CMM (micro in my case) and is distributed to a cluster of let’s say 6 APs. Each AP can pass 7Mb of throughput through it (without 2x turned on in hardware scheduling) right? Does this mean that the cluster needs 42Mb of throughput feeding it? 7Mb for each AP?

I know that from the AP to each SM the throughput is shared. So even though the AP can pass 7Mb, 200 SMs are not going to receive 7Mb all at the same time, correct? Let’s say that each of these 200 SMs are set for 6Mb down 1Mb up. Am I technically oversubscribing my bandwidth after 2 SMs?

Sorry if I made this sound confusing. I’m still trying to understand it all myself.

Here’s my understanding:

First some assumptions:
1) lets assume that you have plenty of bandwidth to your router/cluster manager.
2) lets assume you have 100MBps, full duplex Ethernet connections from your router to the cluster manager and from your cluster manager to the APs.

Each AP does have a theoritical max throughput limitation. But I do not bieleve it is a hard 7mbps. It is based more on the computational abilities of the AP (IE: packets per second, number of SMs connected, etc). As more and more SMs register to an AP, the AP has to spend more and more CPU cycles (and air time (IE: control blocks) ) managing those SMs; thusly there are fewer CPU cycles to devote to packet throughput.

Also, each SM has a dedicated amount of throughput from the SM to the AP. This is what the SM gets throttled to in Prizm/Bam or the SMs config itself. The only way that this “link in the chain” can get reduced is if the AP is running at capacity (according to the paragraph above) and it does not have the resources to handle the traffic.

So, the Internet-bound traffic of all the SMs of all the APs combine in the cluster manager’s switch and head to the router and “out the door.”

In this senario several links are most likely being over-sold. First and foremost is the connection to the Internet. Say you sprung for a full DS3 to the Internet, and it terminates at your tower base (probably not the case, but this is all hypothetical right?). You’ve got auctual throughput of 36mbps. Using your customer plans of 6mbps connections, you’re overselling at customer number 7.

Will 6 customers max out your DS3?

This is the game that all ISPs play. The trick is to know your potiental bottlenecks and monitor them closly (and logically expand them as needed).

If you do it right, you upgrade your next bottleneck at the last possible second (right before it starts affecting customer performance). That way, you’re making the most money possible (by not wasting it paying for performance you don’t need) but you’re not overselling to the point that customer’s notice and get mad (and leave).

You have to look at every point of your network like that: the capacity of your leased lines and your APs all the way down the processing power of your DNS and mail servers.

This is getting really long so I’ll quit boring anyone who may still be reading at this point.

Does this make sence?

Yes, that did help. Thank you Andrew :slight_smile:

Since many Internet users are still on dialup, most websites are built for 128k-256k Internet connections. There are of course content-rich sites that are designed for a fast connection but remember that fast Internet is considered anything over 256k.

Because TCP/IP is not connecting web clients to web servers at the max speed initially, (most of the time the page is loaded before you even hit 500k) and you will never have all customers downloading at the same time, you can oversubscribe a connection more than you realize.

We had 80 customers on a single T1 before it saturated (When it saturated things went to hell pretty fast.) We now have a 10M connection with 138 customers running anywhere from 1M to 5M connections and the router rarely exceeds 5M.

You can further maximize your bandwidth by offering tiered services, and utilizing bursting on the SM’s.

intersting stuff on the size of the web pages…

so in theory if you have customers at 512 and then give them 1MB it should make very little difference to both you and the customer when it comes to general browsing…

Depends on the responsiveness of your network. If it’s clogged with P2P and other junk traffic that won’t fly, but if it’s very clean and low latency you will find that most customers would be pretty happy with 512/1M or 768/1.5M as long as it’s consistent. They just want the pages to load fast.

Of course your power users or your speed test weenies are going to cry about the speed. Increasing bursting should take care of that.

I shape the P2P on to their own link they can knock themselves out…
and the burst deal with the speed tests… so when they complain I just say that it the p2p software… most don’t have a clue…

I had run into problems with P2P bittorrent parallel connections bogging down some current non-Motorola 2.4FHSS equipment. I implemented filtering limiting non “normal” web traffic to X amount of parallel connections and that has seemed to resolve the issue.

I have a question regarding the bursting.

If a customer is set to say 256/128 with a burst of up to 512/256, I understand that the customer can only obtain the burst speed if it’s available. If during the time it is available another customer decides to utilize their connection potential, would customer 1 be reduced back to the base of 256/128? Or would a struggle happen? Or, would the burst be lowered to level of availability?

Customer1: 256/128 base with 512/256 burst
Customer2: 256/128 base with 512/256 burst

If customer1 is utilizing his max speed + burst and customer2 comes into play and uses only 1/2 the connection will that affect the burst of customer1?

I probably confused most of you with my thinking, but I too am confused about bursting speeds.


the burst is not a speed as kps it is just an amount of data i.e 256kb

if your setting is 256/128 and you have a burst of 256/128 that does not mean that the customer will get 512/256 even if it is available… it means the customer will get 256/128 are a rate but is allowed a burst of 256/128 kb. so if there is nothing on the network then for 1 sec you may get 512/256 and then it will drop down, or you may get 256 spread over 10 seconds hence an extra 25.6/12.8 kps for 10 seconds and then back to 256/128