Wireless Access

Hello everyone

I got a client which have notice that for example he is sitting on his desk which is really near an AP  and well he is connected at 300mbps and then it goes to 200, then 240 then 130 then 300 and so on(with full 5 bar signal)... and he was asking me why is doing that?  i actually never paid attention to that ever before...

Also there is another laptop which is a 1x1 abgn card  which is not that far from the AP.  it has well  3 bars windows signal and its connecting to 13mbs then it goes to 20mbs then 28mbs then it goes again to 13mbs... and so on...

I tried another laptop of him on the same place of the 1x1 abgn card which is better... of course because i think its a 2x2:2 card... the thing is that he is connecting to 300 for like 10 mins and then drops to 130mbps

The client is telling me that it didnt happen before... but i honestly think that he didnt notice it before(but i could be wrong)... and well he doesn tlike it.

You actuallly have to be in from of the AP to get a stable 300mbps and for example in the 1x1 card a stable 65mbps

It this a normal behavior?i ask this because like i said i never put attention to this before neither had a client asking me about that...

Config well

They are all APS 105

Band steering

airtime fairness

drop broadcast and multicast

encryptation is on AES

3 SSIDS(will reduce it to 2 when he gets the NPS)

Link aggregations everywhere

Clients are with the lastest driver i could find.

I honestly don thtink  its a config issue...

Neither there are many clients, on the AP, i was testing with just 3 clients

1 1x1  laptop

2  2x2:2 laptops.

Any idea what i could try to make it more stable? or  like i asked before is this normal? what does make this happen?

ideas are welcome :)

Anyone?

The most definitive way to find out why is to do a packet capture while the client is doing that.  Clients will negotiate up or down based on their ability to successfully transmit a frame.  If there is more congestion/contention the client will negotiate down.

What I do not know is if what the Windows supplicant reflects as  the speed is truly accurate.  A packet capture will determine what is really going on.

Windows doesn't reflect rates well in my experience. It's fairly ham-fisted with it.

I generally tell customers to ignore this and try something "real" (upload/download something) to get satisfaction.

Packet capture is a good idea. Maybe as another idea...

If the rest of the environment is as you describe, I generally suggest working with a "known good" client first. This is fairly quick.

For instance, I've carried my latest Atheros client machine (which runs like a train now I've tweaked a few settings) to every install in the last 3 years. In the last couple of scenarios like this, I park my laptop near the "trouble" client to demonsrate the AP/controller estate is ok. It's quick and dirty to prove things this way. Maybe try something similar with a device of yours you know is solid.

If it's not, go packet capturing maybe?

Yep that sounds like a good idea 

Guess i need to take somethings in mind when telling to the client

For example that wireless is a shared medium when wired is not.

there is a half of the throughput that need to take in consideration

So how much woudl you expect of speed if you are connected for example at 300Mbps?

Guess that would be maybe 300Mbps /2 = 150 Mbps

Then making the convertion  150Mbps/8 = 18.75MB/sec

That is what im expecting to see on the data file transfer if im doing a FTP transfer? or it would be less?

Take in mind that the test is done when im the only client connected to the network on a 5ghz band, really near the ap  with 40mhz channel and a client that support all that.

Generally speaking, you're on the right lines.

However, the only thing I would say is that an 11n client consistently going over 18Mbyte is rare. The potential for this is generally in the hands of the client, in terms of adapter and antenna and driver quality (and OS to some extent).

For example, I've tuned my Atheros card as much as I can (it's a good adapter), and the most I can crank out of it is 18.1Mbyte. Iperf/Jperf and Chariot are good ways to test real throughput (on a couple of controlled machines).

Unless you want to spend a long long long long time analysing it, I'd say that anything over 15Mbyte (clean network and RF) in a basic test is a win.

The Racking Monkey,

I would have to disagree with you on the throughput number above.  We have seen much higher throughput for clients, but that is not what Nightshade is talking about.  He is talking about the association rate, which is the speed that a client negotiates with an access point at that tim.

Let us remember that this thread is about finding out why the client reflect the association rate it does, NOT throughput. For that if  we simply grab a packet capture while the client is doing that, there is a column that will show the association rate and say if the client is reflecting what is truly in the air.

Understood, but it looks to me like Nightshade has then started discussing actual throughput to me (Mbytes and FTP)?

Absolutely, when looking at association and throughput, focus should be in different places.

Nightshade will advise what his primary interest is I'm sure?

Regarding client potential, that's not what I'd say we see in the UK. Maybe it's different in the US (due to density or geography to some degree). It's very rare over here to see a client (for instance a HP laptop) get over 18Mbyte (when associated at 300Mbit). In the last 3 years, I think I've probably seen 4 or 5 good factory build corporate laptops. And I've seen countless (more than 50) perform less well than you'd ideally like.

My interest is both i actually want to know both  :)

If you want Collin i can open another tread for the other one.  I can also do a packet capture to check that :)

But i have seem in my test this.

When im putting the AP in bridge mode i get the actual trhougput im looking for and what i have calculated... i have seem on my machine 18.5  speed on my FTP server transfer speed.

I think the ranking monkey is right in the way that if i show this the client, he wont care what association speed he see on his laptop... he just thinking hey! why im connected to 108 instead of 300??? im getting lower speed...

But in the test on the AP on bridge mode im on 108mbs connected  i get 18.5mbs  which is the speed of 300mbs

Now if i put the AP in tunnel mode i get a total of 4MB/sec which is a way low...  This inspact is really high in the througput!

Now my scenario looks like this

CORE SWITCH----SWITCH----WC

The AP i plugged in the CORE SWITCH(this is just a testing lab) just because is poe and  is 1 giga.

Anyways all the uplinks are giga, i made sure everything was running on 1 giga...

I got no traffic going to the Switch  that i plugged my WC besides my wireless traffic.

Does the tunnel mode affect this much the throuput? i mean i know it has to tunnel to the  WC... and has to get back... but well  i kneow it affected in some way but not this mucc... its like 4 times slower.

If this is true  it is really important where you place the WC and also its really important to put in link aggregatioons...

Ill try  connecting it directly to the Coreswitch which it should a make  a differnce a think. its one less hop he has to do...  but well just got a controller 620 so i just got one giga port so well no link aggs...

Any commend about this?

Ummmm. Very interesting.

So your bridge test is good? I think the important points to take away from that is...

1. The AP can perform fine RF wise.

2. The client seems ok.

3. Ignore the association rate reported. The OS or driver is probably mis-representing it. Don't worry about it.

Moving on...

Unless something in the controller config is affecting the throughput, you should get more than 4Mbps. That's wrong. So, I'd suggest you...

1. Check what version of code is on the 620. Upgrade it to the latest if possible.

2. If it's easy to move the controller to the core switch, yes do that.

3. Check the controller to switch port for errors and anomalies, plus speed and duplex. Consider fixing it at 1Gbps, as I've seen that help before. Depends on the switch.

4. Is the user ingress VLAN the same with the AP in tunnel mode as when in bridged? If not, try changing the tunnel AP VLAN to the same as the one in the bridge test. Any improvement?

5. Log a TAC case???

Good luck!

Yeah i did more test...

I moved my controller and plugged it directly to the CORE SWITCH like i always do on the cilents...

The throupgut speed improved from 3-4MB/sec to 10-11MB/sec which is  a greate improve but still is slower than when i put bridge mode.

I already checked if i had errors on the ports which i don t have no anomalies no nothing... its all good.

In tunnel mode all the vlans are separatedly....  the Wireless vlan is another than the wired vlan... on the case on the bridge more its the same... because i just wanted a fast test...  and i think it wont affect that much i mean... on that wired vlan there is just one machine for now... and also i got drop broadcast and multicast...

Anyways moving on wow this things are things that we should also take in consideration when desigining this solution... i didnt realize how important it is... i have been doing it to fallow like the best practice.. but didnt realize how much differnet it does....

Now i guess like if i would be able to put the link aggregation iw ould able to achive those 18.5MB/sec im getting on Bridge mode on tunnel mode.

I honestly don't think it will be to do with you needing to aggregate more ports into the controller. The port you've got attached is gigabit yes?

What version of code is running on your 620?

Yep of course is on Gigabit  i already checked that is really syncronized to 1gb no erros blah blah blha. :)

The version is 6.2.0.3

Why you honestly don tthink it will improve the performance?

By now the performance is on top when puttting it on brdige

On tunnel mode in the switch on my desk its on 4MB/sec

When plugging the Controller directly to the core improved to 10MB/sec

Dont think is the switch on my desk..  as if i plug my computer directly a good throughput...  now we have to be really clear what is the best way to connect everything as it clearley  affect the throughput... or at least it what im looking on my lab...

Dont have a 650 to test a linkagg to see if it improve...

You're not going to get anywhere close to 1Gbit throughput in your test, so I don't think adding extra ports to the controller-switch connection is going to add value at this stage.

I can't explain the difference you see when attaching to different switches without seeing a network topology diagram, including switch models and configs.

First up though, 6.2.0.3 is an early deployment code. If I were you, I'd try downgrading to 6.1.3.7 and re-testing.

Of course it wont get any near 1 gb i know that.  In a real scenario this would make a differnece... in my lab it wont.   As the maximum teorical throughput its 600... 300 for each band.. i know that...

I mean when you get other things like more APS  traffic in the network... Which is the client  scenario...

In my lab Im getting 18.5 and its GREAT  its what i expect......

now thing is that those were the results

Bridge mode = total throughput i expect

AP on tunel mode  like 14  which is less...

Naturally the Traffic goes from the AP to the wireless controller then to the CORE switch, then it send it to the destination, then it send it back to the core switch and then it send it to the controller and then it send it back inside the tunnel bakc to the AP...

Now im loosing throughput doing all that...

I just don twant you get the worng im idea im lost in this... i know what you mean when you telling me the thorughput wont get near close the 1 giga i just wasnt thinking when i said it in a single AP and a single client, i was thinking in the real enviroment which is like 16 APS.   All APS going to the maximum can bottle neck the single gigabit interface between the controller and the switchore... it was just a misunderstanding though.

Now my real questoin would be

What snould be the throughput i should expect from a AP that is tunneling to the controller?

I guess the asnwer would be that it totally depends...

Depends that all the uplinks are giga... and much better if they are doing LAGG.  How many other traffic is going through that single interface that goes from the switch that the AP is plugged to the switch core?

But when i did the test all the interfeces between the AP and the Wireless controller which i had something liek this

WC---SWCORE--SWL2----AP

You actually have to take in mind many things when deploying the APS, in which Switches you will put it... as it could be a 10/100/1000 48 ports switch  which is plugged with just 1 giga internface to the core... if for some reason the other ports start doing a file transfer  whcih  flood that link between the SWCORE And the SWL2 then your througput is going down... anwyay seven if you put it on bridge mode... thats why i suggest putting LAGG everywhere i can.

 Plus the downgraade i already tested it and the results were the same.

At the end on the client even if the throughput was around  12-15 which is great, he still saw the windows thing doing difference speed on the client  which he can see and he didnt liked it.... at the end he just got another AP inside his office.... now he does not see it and now he is happy...

I just keep the discussion open becasuse well its interesing...

And from this i learn that i should just ignore what the windows says...

Ok, cool.

Theoretically, I wouldn't expect the tunnel throughput to be much different to bridged in a clean test with 1 client.

There is definately something strange/busy going on within the wired network, as you said moving the AP to the core had an improvement on tunnel performance.

Don't know what to suggest without looking at configs on switches and utilisation of them.

Best of luck!

Ill have to take a look to the config of the switch more in deep... it just that, that switch got a lot of things configured...