Forum

Hey Guys,

I wanted to pick your brain on an 802.11n solution that I am evaluating, and see if you can provide some additional insight.

We are in a development phase for an 802.11n implementation here the University of Colorado.
We are moving forward with 802.11n in the 5GHz spectrum using 40MHz channels MCS 15 primarily on Cisco 1142 AP???¡é?¡é?????¡é???¡és.

As you can imagine we have a plethora of wireless clients on campus. In testing we also have 802.11n enabled in 2.4GHz using 20MHz channels.
I???¡é?¡é?????¡é???¡éve done this with the intent to accommodate backwards compatibility of legacy clients in the 2.4 band, and to have them reap the benefits of faster data rates/resiliency/range/and predictability that 802.11n provides.

Even with Cisco Band Select enabled, I still see some 802.11n clients move to the 2.4GHz spectrum albeit at 100Mbps-150Mbps. This is acceptable, however I am not sure how 802.11n Protection Modes are effecting the performance of both legacy and 802.11n clients. I???¡é?¡é?????¡é???¡ém waiting until we have a broader client base before we can truly test this.

I want to see what comments and recommendations anyone might have regarding this configuration.
I appreciate any information you might be able to provide.

Thanks,
-Max Lopez

mrmax Escribi?3:

Hey Guys,


As you can imagine we have a plethora of wireless clients on campus. In testing we also have 802.11n enabled in 2.4GHz using 20MHz channels.
I???¡é?¡é?????¡é???¡éve done this with the intent to accommodate backwards compatibility of legacy clients in the 2.4 band, and to have them reap the benefits of faster data rates/resiliency/range/and predictability that 802.11n provides.

Thanks,
-Max Lopez

If You want to increase Your .11n network speed, You must move to 40 MHz cahnnels in 5 GHz band. Use 40 MHz channels in 2.4 band isn't good idea. Look for more details http://www.youtube.com/gthillwifi#p/a/u/0/YNYM1Gn0Rzg

Vertigo

Vertigo,

Thanks for the reference to my video. :)

Max,

I'm assuming that Cisco Band Select is their version of band steering. The way band steering works isn't perfect so you will see a dual band device associate to the 2.4 GHz side of the AP. This can happen for a number of reasons, depending on the algorithm that Cisco uses. Sometimes having too low of a signal on 5 GHz vs. 2.4 GHz can cause this, but that is way out there speculation.

You could also employ psychological band steering. Run dual SSID's, one per radio. The 2.4 GHz SSID is UoCWiFi and the 5 GHz is UoCWiFiFast. (I'm stealing this idea from Keith Parsons). That way if a student's computer is dual band it will show the "fast" SSID and hopefully they choose that one.

Having 11n devices on the 2.4 GHz radio is fine. You just run normal 1,6,11 channels without channel bonding. If some 11n devices end up there, it's no problem. Since you will have a multitude of legacy devices it will slow down vs. 11n only, but there is nothing you can do about that. The legacy people will actually have better performance than they would have if you had 11a/g AP's.

Not sure if I have fully answered your questions, but feel free to shout back if I can help.

GT

Hey GT,

Yes Band Select is Cisco's flavor of band steering. I expected to see dual band clients drop to the 2.4GHz side even with Band Select enabled.

It's an interesting configuration you mention with dual SSID's. We currently have the same SSID on both radios. We are looking to NOT confuse students too much ;) and accommodate all client types that connect to the network. This has worked well for us in past with 802.11a/b/g clients. We essentially allow students to connect to a universal SSID and have their client negotiate the best frequency and data rates that are available on their respective AP.

With 802.11n, I wanted to use this same philosophy however I was unsure how an 802.11n client would behave with an 802.11g client using a 20MHz <unbonded> channel in the 2.4GHz range. My goal is to have legacy clients reap the benefits that 802.11n can offer them. However knowing that 802.11n clients will also be sharing the medium, I wasn't sure how the Protection modes would kick in for 802.11n. Would 802.11n clients drop down to 802.11g speeds at that point?

You did defiantly give me comfort in knowing that these clients can co-exist on an unbonded 2.4GHz 802.11n network.

I appreciate the feedback.
Thanks,
-Max

Not exactly the same, but some interesting stuff in the same vein nevertheless:

https://supportforums.cisco.com/thread/343687.pdf;jsessionid=53A2CA3BB2E6EFCDB8DE8AB352AAA103.node0

Dave

mrmax

802.11n has a very complex set of protection mechanisms. I won???¡é?¡é?????¡é???¡ét go into all the details, but there are essentially four protection types called Protection Modes. These are numbered 0-3.

Mode 0 is Greenfield Mode, where everybody is operating as a High Throughput [ HT ] station. You can have either of the following:

A. Everybody is operating on 20 Mhz B/W only [ AP???¡é?¡é?????¡é???¡és, clients ]. If any stations are detected on the channel, then they must also be HT stations. That is, no .11g???¡é?¡é?????¡é???¡és etc.

B. Everybody is 20/40 Mhz capable. If any stations are detected on the channel, then they must also be HT stations. That is, no .11g???¡é?¡é?????¡é???¡és etc.

It is called Greenfield mode from the real estate developer term when you have a physical green field [ all green grass, no nasty weeds or anything to blight the symmetry ]. If everybody is working like this, then we do not need protection. You have to be very, very lucky to have this scenario. In this case we have HT members with the same channel width and HT non-members [ of the BSS ].

Mode 1 is called HT Non-Member Protection Mode. Here, we are working all HT as before, either pure 20 Mhz or 20/40 Mhz capable. The ???¡é?¡é?????¡­?¡°nuisance???¡é?¡é???????? here is that if a non-HT station is detected in the primary or secondary channel, then protection mechanisms must kick in according to the .11n spec. Even if just one STA is heard, those mechanisms must kick in. The .11n guys don???¡é?¡é?????¡é???¡ét actually drop down to .11g speeds, but in some cases they might as well do. What happens is an overall slowing occurs, as say .11g STA???¡é?¡é?????¡é???¡és take longer to transmit the same amount of data that a .11n STA would. That in conjunction with the CTS-Self or other actual protection method overhead [ as well as other stuff ] cause an overall perceived slowdown. In .11g/.11b protection, it is similar. The .11g guys do not actually drop to .11b rates, but again there is an overall average reduction in throughput. In this case we have HT members with possibly different channel widths [ e.g. 20/40 Mhz BSS with either 20/40 Mhz HT station members or 20 Mhz HT station members ] and at least one non-HT station non-member has been detected in the primary or secondary channel.

Mode 3 is called HT 20 Mhz Protection Mode. This is ???¡é?¡é?????¡­?¡°Son of Greenfield???¡é?¡é????????. In this one, we have a bunch of HT stations working 20/40 Mhz. Now, we have one or more STA???¡é?¡é?????¡é???¡és only capable of working 20 Mhz associated to the AP. Protection kicks in. This one is referred to as HT members with possibly different channel widths and HT non-members.

Mode 4 is called Mixed Mode. Here, we have either a 20 Mhz only set-up or a 20/40Mhz setup. This time, however, one or more non-HT stations are actually associated with the AP [ called a ???¡é?¡é?????¡­?¡°member???¡é?¡é???????? ]. Protection kicks in once more with this one. Referred to as HT and non-HT members.

On 5 Ghz, Mode 1 wouldn???¡é?¡é?????¡é???¡ét be that common, as we are really only talking about .11a sticking it???¡é?¡é?????¡é???¡és nose in here, and there aren???¡é?¡é?????¡é???¡ét too many of those folks around [ in general ], although in some areas that may not be the case.

Scanning and channel management [ Extended Channel Switch Announcements etc ] are other important areas.

In summary ???¡é?¡é???????|.Let???¡é?¡é?????¡é???¡és find who thought this stuff up and pay them a visit ???¡é?¡é???????|..

Dave

When 802.11n was being ???¡é?¡é?????¡­?¡°designed???¡é?¡é????????, the ???¡é?¡é?????¡­?¡°designers???¡é?¡é???????? were adamant about one thing: even though .11n was going to be this super-dooper high speed system, it should have no more or no less ???¡é?¡é?????¡­?¡°rights???¡é?¡é???????? on the airwaves than existing pre-.11n systems as neoghbours. This is why the protection mechanism specifications are fairly complex. Also, within .11n itself, a 40 Mhz capable STA should have no more or no less ???¡é?¡é?????¡­?¡°rights???¡é?¡é???????? than a 20 Mz [ only ] capable STA. It???¡é?¡é?????¡é???¡és not perfect, but it does a pretty good job.

Unless you are very lucky and have a clean ???¡é?¡é?????¡­?¡°Greenfield???¡é?¡é????????, most folks on 2.4 Gigs are going to have ???¡é?¡é?????¡­?¡°mixed-mode???¡é?¡é???????? systems where 20 Mhz .11n systems have to co-exist with .11g and even .11b systems.

So where does the myth that .11n systems drop down to .11g speeds [ in the presence of .11g systems ] come from ?

Firstly, in a mixed mode environment, right down at the physical layer, we need a mechanism to allow .11g [ just ignoring .11b for the sake of repeating ] systems to recognize that a .11n transmission is taking place. In a Greenfield environment, we can use an HT Greenfield PPDU which can only be understood by other HT Greenfield STA???¡é?¡é?????¡é???¡és. This PPDU has a physical header of a certain length. For a mixed format system, we need to ???¡é?¡é?????¡­?¡°add on???¡é?¡é???????? another header prior to our ???¡é?¡é?????¡­?¡°main???¡é?¡é???????? .11n header so that legacy systems can understand that a .11n transmission is going to take place and for how long. The legacy system does not have to understand anything after the legacy header portion and in fact it physically cannot ] All it has to do is obey the information in the Legacy Signalling [ L-Sig ] portion of the header, which basically says ???¡é?¡é?????¡­?¡° Hey, you have to hold off trying to transmit until the .11n sytem is finished, according to the time given in the Sig portion???¡é?¡é????????. In actual fact, it doesn???¡é?¡é?????¡é???¡ét give a direct time period, but the microprocessor can figure it out using the L-Length [ Legacy Length ] and L-DataRate [ Legacy Data rate ] value in the Sig Field. .11n is nothing if not fair, so even at this stage, they said what should be the L-Data Rate value ? ???¡é?¡é?????¡­?¡°???¡é?¡é???????? They used the lowest value that is ???¡é?¡é?????¡­?¡°specd???¡é?¡é???????? for OFDM for .11g and .11a which happens to be 6 Mbps [ from the old basic ???¡é?¡é?????¡­?¡°must have???¡é?¡é???????? set of 6, 12, 24 Mbps ]. So this added overhead on top of the ???¡é?¡é?????¡­?¡°true???¡é?¡é???????? .11n overhead is the first ???¡é?¡é?????¡­?¡°slowing down effect???¡é?¡é????????.

Secondly, we need to have a ???¡é?¡é?????¡­?¡°warning???¡é?¡é???????? to legacy systems that a .11n transmission need s to occur, and that is where the true protection mechanism comes in. Methods such as CTS to Self are used, [ RTS also comes into play ] in a dual mode, where one CTS-Self is sent at the lowest supported rate of the legacy system [ if .11b is present, that means it could be sent at 1 Mbps !! due to the fact that the ???¡é?¡é?????¡­?¡°b???¡é?¡é???????? spec says the supported rates are 1, 2, 5.5 and 11 Mbps ]. This is all because .11n says ???¡é?¡é?????¡­?¡°If we are going to be fair, we must be fair to everyone, including the slowest of the slow???¡é?¡é????????. Two problems occur here. Firstly these are ???¡é?¡é?????¡­?¡°extra things???¡é?¡é???????? which require physical time to transmit [ and even then, some are at very slow rates ]. Secondly, these frames have to ???¡é?¡é?????¡­?¡°compete???¡é?¡é???????? under the rules of CSMA/CA just like any others [ CCA and all that good stuff ].

Thirdly, when a legacy station transmits [ at 11 Mbps, 24 Mbps or whatever ], the .11n STA just has to ???¡é?¡é?????¡­?¡°bite it???¡é?¡é?????¡é???¡és tongue???¡é?¡é???????? and wait until the [ to it ] frustratingly slow .11g or b STA finishes it???¡é?¡é?????¡é???¡és transmission.

When the .11n STA actually transmits, it is not at .11g or .11b speeds. It is at the actual .11n speed appopriate to it???¡é?¡é?????¡é???¡és MCS [ Modulation and Coding Scheme ] etc etc. The reduction in throughput from the .11n system point of view is due to all the stuff mentioned above.

This reduction can vary dramatically. Suppose that we had a bunch of .11n???¡é?¡é?????¡é???¡és and one single .11g associated with an AP.

Suppose that the owner of the .11g STA using the AP only used it for checking his e-mail infrequently [ not very likely, but useful as an example ]. Actual traffic effects on the overall .11n would not be that high. However , provided he is still associated, we still have an HT Protection Mode situation going on, and the .11n???¡é?¡é?????¡é???¡és still have to ???¡é?¡é?????¡­?¡°send out warning shots???¡é?¡é???????? that they are about to transmit.

In a real situation, you are more likely to have a bunch of folks on .11g transmitting at all sorts of times and with varying amounts of data. All will still have to contend in a CSMA/CA manner.

There are a lot of complex factors going on, but one thing remains constant ???¡é?¡é?????¡é?€?? just as in 802.11g/b protection, the .11n folks do not drop their physical transmission rates down to .11g speeds. It???¡é?¡é?????¡é???¡és just because of all the stuff mentioned previously that the reduction in throughput [ compared to Greenfield ] makes it appear that way.

Dave

Bye the way, you might be saying to yourself ???¡é?¡é?????¡­?¡°Why do we even need CTS-Self etc when in Mixed-Mode [ Legacy Mode ] we already have that ???¡é?¡é?????¡­?¡°extra???¡é?¡é???????? legacy readable header ???¡é?¡é?????¡­?¡°tacked on???¡é?¡é???????? to the .11n readable header ? Surely we now have the warning that a 11.n transmission will take place ????¡é?¡é????????

This is true, but there are some other reasons that make us use CTS etc. Won???¡é?¡é?????¡é???¡ét go into all of them here, but one of them is to do with how the physical layer of your card actually recognizes and ???¡é?¡é?????¡­?¡°locks-onto???¡é?¡é????????a physical header. Two processes called carrier recovery and bit-timing recovery occur. Associated with this is an ???¡é?¡é?????¡­?¡°acquistion process???¡é?¡é????????. Historically, designers have designed this so that the smallest expected Inter-Frame space must be accomodated. Traditionally, this has been the good old SIFS. However with the introduction of .11n, a new space called the RIFS [ Reduced inter-frame space ] may come into play. This guy is even smaller than the SIFS. This could cause all sorts of horrors in terms of correct processing of frames that the .11n designers came up with additional protection mechanisms.

Just in general, they realized that due to the enormous number of possible ???¡é?¡é?????¡­?¡°features???¡é?¡é???????? of .11n, some manufacturers might not choose to implement all of them, and this had to be accommodated for. In other words, ???¡é?¡é?????¡­?¡°not all .11n???¡é?¡é?????¡é???¡és are created equal???¡é?¡é????????.

Dave

A bunch of different protection mechanisms are actually available including one called L-SIG TXOP. Information provided in the HT and ERP information elements are used in the decision of what type will be implemented.

Dave

Dave, you are a fountain of information.

Max, long story short... remember how 11b and 11g coexistence effects 11g? 11g stations don't get 11b rates, but the performance is definitely impacted. Expect the same thing for 11n. You'll still see better performance with 11n clients than 11g clients, but not by much, especially in areas with high utilization. Even though 11n STAs can still transmit at higher data rates, you have protection overhead, and faster STAs get hung up a bit because 11g STAs use slower rates. Airtime fairness may be a feature to look into. I know Cisco claims ATF as a part of M-Drive, but I haven't seen it myself, so I can't say what they're doing there. Either way, you can be reassured that 11n among legacy clients in 2.4 GHz will still see a slight performance boost. In your environment at the University, Band Steering and Air-Time Fairness are probably the two best features working for you with all those legacy clients around. Now that I think of it, ClientLink should help too by maximizing downlink (most traffic) performance to legacy clients.