Wireless laptop roaming through various access points

you've mentioned Ralink RT2500 have you tried its utility ? Has 'fast switching option' allows you to set roaming signal threshold, default is -70db ... no delay other than an occasional instant hickup rarely noticed

I've got a Dell Latitude D610 with Intel 2200BG and it seems to switch seamlessly between two Linksys WAP54G APs. Try fiddling with the "Roaming agressiveness" knob in the driver...

I have two Linksys WAP54G and one WRT54G. However, SSIDs are different. Will try to give them all the same SSID. Why should it be the same ID? Will that make the rescan quicker, as there is only one AP in the nic's list?

actually I used the Conceptronic utility V 3.0.37.50503 Driver v. 3.00.02.0000 however, I didn't see the "fast switching option". I did se a "Group Roaming" option and it's enabled. Ref.

Are you referring to the same utility or do you suggest I download another? Like from the

Who are you replying to? The Intel 9.0.3.9 driver has some advanced settings in the driver (under Device Mangler) you might want to try...

William,

I previously posted two replies, one for you and one for bumtracks.

Thanks for your feedback. I did try just that software from Intel's site:

Regarding WiFi nic bridging I mentioned in my original post, is it an absurd idea? (haven't tried it yet but I doubt it will work)

OK, and if you had included a snippet of the message you were replying to, it would have made everything clear. Just like I did above... Oh well.

Are you switching between APs with the same SSID? I beleive you have to have the same SSID in order for the roaming thing in the Intel drivers to function properly. I don't remember if I had mine on different channels or the same channel, IIRC there were advantages either way...

Also, FWIW, I was using straight APs (Linksys WRT54G boxes, one V1 and one V{3,4}, with the latest firmware, and not a WRT54G router. They are supposed to be the same, but there are some differences...

That's what I have left to try: set same SSID on all APs. Will try that out tomorrow. However, I suppose channels for "adjacent" APs must be different to minimize interference. But maybe if they don't overlap too much maybe using the same channel has its roaming advantages... or at least picking neighboring channels. Will give it a run.

No, either pick the same channel for all your APs or chose from the set of non-interfering ones (1,6,11). I forget which is 'better', let us know when you do your testing tomorrow.

Here's what I tried:

One of the APs is on channel 14 (2.472GHz) and the other is on channel 9 (2.452GHz)

I used two laptops:

Conclusion: I'm satisfied with the roaming performance of both cards: Intel and Broadcom. However I would like to learn how to fine-tune wireless connectivity. Also, WEP will be insufficient and am working on a freeRadius server which will get its LDAP data from a LAN Windows 2003 Active Directory server. I wonder if roaming will downgrade if I add Radius authentication.

Regarding the channel selection you mention, I will try to use the same channel for all APs (will pick channel 11). I will check whether they interfere or not, even if their radio coverage overlaps a bit. But then which channels should I pick for each client laptop? same as the AP's? Or should they be different from client to client?

Anyway, thank you for your help. The "same SSID" for all APs really helped.

I believe laptop channel selection is for ad-hoc connections only. So in my case I can ignore the client laptop's radio channel number. Please correct me if I'm wrong.

Just leave everything at it's default settings, IMHO there's no point in trying to 'tune' for better performance.

Where are you located?

Again, fiddling with settings is probably going to break something...

If you are satisfied, why not leave it alone?

So turn on WPA.

IIRC, if you only have a few clients, they shouldn't interfere noticably.

The client 'channels' are for ad-hoc mode only, and can be safely ignored in this case. In infrastructure mode the laptops scan all the channels looking for their APs.

in Europe. Yes, I know, we have more channels than in the states...

Actually, the reason why I'm fiddling is that I noticed that the Broadcom card on the HP Compaq nx6125 isn't as "aggressive" as the Intel 2200BG. Thus, if I directly compare, under the same conditions, a Broadcom and an Intel, I can record a 3-5 second "downtime" for the Broadcom while switching APs, and a practically seamless switch for the Intel. I'm still satisfied with both because in neither case does a, e.g., Terminal Service connection break. However, some users can get impatient in even less than 5 seconds... Of course, we could buy exclusively Intel-based laptops but we already have a couple of AMD-Broadcom devices and it's a shame not to use them (especially the brand new AMD Turion 64). So basically, that's why I'm "fine-tuning". Do you have a roaming AMD-Broadcom laptop or just Intel as you mentioned in a previous post?

The unit where I'm going to install the wireless network will be around

-------------------------------------------------------------------------------

- approxiamtely 16 bedrooms -

-------------------------------------------------------------------------------

- 500 feet hall -

- AP1 Ch. 1 AP2 Ch. 6 AP3 Ch. 11-

-------------------------------------------------------------------------------

- small rooms

-------------------------------------------------------------------------------

Did I understand you correctly regarding channels 1,6,11? Also, the "best coverage" should be for the 16 bedrooms (each is 15 feet deep, max.). The "small rooms" actually don't need excellent signals although it would be nice to have coverage there too. So I was thinking of mounting the APs vertically on the top quarter of the hall wall as described in the ascii drawing above, so that RF signals bounce off of that wall and direct signals mainly towards the bedrooms but would have enough remaining for the nearby "small rooms" (or at least that's my "theory"). Or do you think it would be better to put the APs on the other side of the hall, i.e. on the wall right next to the bedrooms? (the APs can't be mounted upside-down in the center of the hall).

As a side note, I also tried the following:

on a single laptop I used two wireless cards and bridged them. One card was configured to link to an AP with SSID AP1 Channel 1 and the second card was configured to link to an AP with SSID AP2 Channel 6. It seems that the bridge actually allowed the laptop to preserve its DHCP-assigned IP throughout roaming. So that part was ok. However, I thought that if I had two cards, both would try to connect to the APs on their respective lists *at any given time*. They *do* connect correctly but it seems that while one card's link is up, the second doesn't try to connect to *its* AP (if it did, then as soon as the first card's link goes down, the second would already be up and running).

So roaming actually "worked" in a bridged dual wifi nic laptop but the downtime was unacceptably long (10 seconds).

Has anyone tried bridging wireless cards?

William P.N. Smith hath wroth:

Not exactly. All the multiple interfaces on my office machines (two ethernet, wireless, Bluetooth, dialup modem, VPN, etc, are alive and well at the same time. What's different is that MS can normally handle only ONE default route (or default gateway) for internet traffic. Whichever interface has the default route pointing to it, gets all the internet traffic. I have several static routes pointing to remote networks (i.e. VPN's) that direct some traffic to specific destinations, but the internet traffic usually goes via the default route. It is possible to assign more than one default gateway in the configuration for a given network interface, which is useful for load balancing, but usually does weird things because there's no RIP protocol to redirect the traffic to the correct gateway.

What Windoze is suppose to do when juggling enabled and disabled interfaces is to change the default route. This is usually not a problem with the common wireless router because the IP address of gateway is identical whether it's going via a wired or wireless bridged link. Same thing with multiple access points connected to a common gateway machine.

What needs to change is the ARP table, which maps the MAC address to the IP address. The IP address doesn't change but the MAC address of the access point should. In theory, spanning tree protocol (802.1D) takes care of lost links at the MAC layer. In reality, it tends to be a bit chatty and is rarely implimented in wireless bridges. It's just not fast enough to deal with moving target networks. The new 802.11r, which includes a fast version of STP, should solve the problem.

Meanwhile, if you just monitor the ARP table with: arp -a you can see if the client is switching between access points (i.e. betweeen MAC addresses) as the MAC address of the default gateway will change.

I beleive this is a function of how WinDoze handles having more than one network interface (it ignores all but one of them).

I've managed to roam seamlessly both with Broadcom and Intel wireless laptops. However, this is true for open APs or with WEP encryption. If I use WPA-Enterprise with a Radius server, there's a long delay when switching between APs.

The test environment is as follows: freeRadius.org Gentoo Linux server --- 3 Linksys WAP54G APs configured with WPA-Enterprise w/Radius --- 1 roaming laptop

The delay seems to be due to re-authentication with the freeRadius server and that seems to be "expensive".

Is there a way of "caching" or "pre-authenticating" or "propagating authentication between APs"? Cisco and Proxim products seem to do this but I would like to do something "acceptable" with cheaper APs such as the Linksys products.

Has anyone found a solution to this roaming problem?

Isn't that the forthcoming 802.11r(?) stuff for fast roaming? IIRC, it keeps it's connection to the old AP while it gets reauthorized thru the new one, then switches seamlessly. It's not ratified yet, so don't expect it for a while, and it may (*) require new hardware...

(*) or may not. "We _demand_ rigidly defined areas of doubt and uncertainty!"

"VDP" hath wroth:

The limiting factor is not RADIUS pre-authentication but rather how "sticky" the client software on your laptop wants to stay connected to the old access point. There is no provision or decision mechanism for switching access points in most clients other than total loss of signal or de-authentication. See: |

How long a delay? It takes 800-3000 msec to setup new association, authenticate, and switch WPA keys. Double that to de-authenticate from the old access point.

Not that I know about. It's one connection at a time, period. However, there's no technical reason that a bridge cannot associate with more than one access point at a time. The major limitation is that clients are hard coded to bridge exactly one MAC address at a time, thereby preventing additional simultaneous connections.

I think 802.11r is not really needed as 802.11k addresses pre-authentication and 802.11i addresses fast 20-30msec connections and also pre-authentication. See:

They have their own protocols to do roaming. So does Symbol. The "wireless switch" vendors also have such a features, where having all the intelligence centralized in one box makes it easy.

Netgear re-sells AutoCell:

Only with single vendor equipment. There's little that can be added to existing random hardware and software to solve the problem. Intel Proset works well enough switching in about 3 seconds, but I've never tried it with a RADIUS server.