Hi this might be a n00b question but I'm about to create a link between two building's networks and what I have planned is:

              PC                                                                              multiple wireless clients
                |                                                                                               |
WAN <-->AP1 <----WDS-----> AP2 <-----WDS-----> AP3 <------WDS------> AP4
              -------------------------------                      ----------------------------------
                       bldg 1                            300m                         bldg 2

Which kind of seems like a waste - if I understand correctly every WDS link halves the bandwidth of the channel so I'll have 54/2/2/2/2 =~ 3Mbs give or take between all links?
The reason for two APs in each building is that AP1 with the WAN link and the PC is in the basement at the west end of bldg1, AP2 with the directional antenna is on the east end of the 2nd floor, AP3 is in the attic of bldg2, and AP4 has a ton of laptops connected wirelessly on the ground floor...

Anyway, is there some way to do links between OpenWRT units WITHOUT resorting to WDS?

You want to network all of your Aps but you don't want to use WDS? Run a cat5 cable.

WDS isn't inherently problematic, it's really just a point to point link.

The problem arises when you start to do things like bridging between your wireless network and your wired network -- suddenly all the broadcast traffic from your wired network is now taking up air time on your wireless network; add to that several devices linked only via WDS and you get a sequence of "recieve, transmit, recieve, transmit .." as each packet hops across several WDS links. This retransmission is what consumes bandwidth, not WDS itself.

That makes sense. I guess another way to ask my question was "is it possible for an OpenWRT device to be a wireless client and a wireless AP simultaneously" which does, in fact, make it a n00b question. =-)

Thanks.

WDS doesn't slow down by half every step of the link.  Once your data is on a segment not in range of other segments  it doesn't slow it down by the same factor.   From your example AP1 should be able to transmit to AP2 at the same time AP4 is transmitting AP3.  AP2 and AP3 will be the biggest limiter of speed since the see each other and the next one in the chain.   

The best way to set this kind of network up with WDS is to use a routed link instead of bridged.   (See the wiki for some more info on this.)  This will keep the broadcast traffic to a minimum and avoid loops in the WDS network. 

The only way the device can be "client" and AP is with WDS.  Setting up WDS in a routed mode can be more work but worth it in the end.  You can make your wireless network as large and complex as you want this way.  If you really need the speed you need multiple radios at each point.  You could go as far as 3 at each spot, one for uplink, one for downlink and one for local clients.  Also you need to keep these three radios on non overlapping channels (1,6,11).   It's really a matter of how much time and money you want to throw at it. 

Hope this helps more than confuses.

I'll just add one more piece of trivia:

WDS is traditionally set up with each AP on the same channel, because they all need to communicate with each other. But, because the channels overlap, they don't actually need to be on the same channel. What this means is that it's possible to do things like:

AP(ch:1) <--> AP(ch:3) <--> AP(ch:5) <--> AP(ch:7)

Yes, it actually does work. Admittedly it's just a quirk of the channel assignments -- meaning that you will get progressively weaker signals as you move away from the channel the device is transmitting on.

Does WDS on OpenWRT work with WPA-PSK?

Yes.

It's also possible to have unique keys for each wds link, although setting that up can be extremely tricky.

This is quite helpful, yes... I had forgotten that WDS is a MAC layer link, so that bridging and routing are independent of the WDS. So if I break the bridging between the two sites and just route traffic over it I should avoid most of the "crosstalk"

Thank you!!

mbm

What do you mean with unique key. I know you can use static wep key but I have read that with wds you cannot use dynamic keys like wpa-psk.
Mostly because of this I don't use wds ...

WDS can use wpa keys, although you may find that non-openwrt devices don't support wpa on WDS. There should be more information on this elsewhere in the forum (I don't have time to go digging for it -- try searching for "wl0_wds_wpa_psk").

Not only did that work, it seemed the best config for my setup -- all three on one channel was giving me ~20% packet loss. I spaced out the channels in the way you suggested and it improved immensely - down to ~3% loss.

BTW, there is an ambiguity in the wiki and other places where I found WDS configurations discussed which caused me some confusion and consternation -- I originally added the MAC addresses of *all* of the the APs to each AP - excluding itself, of course. This made them very confused - pings got "DUP!" responses and all IP would eventually grind to a total halt. I fixed this by only giving the MAC address of the "middle" AP to each "edge" AP and giving the two "edge" APs to the "middle" AP. It seemed that linking the two "edge" APs made a circular route at the link layer and b0rked everything... or am I imagining things?

The wl0_wds variable is the list of peers to connect to, if you only have two APs then wl0_wds on each gets set to the mac of the other. When you have multiple WDS clients in a bridge, things get more interesting. If you had 3 APs and set WDS on each to point to the other two you'd end up with this:

    AP
   /  \
  /    \
AP------AP

The bridge forces it to retransmit WDS packets, and you quickly have a storm of duplicates from the retransmissions until your network comes to a grinding halt -- exactly as you described. If you enable spanning tree on the bridge, the loop is detected and the problem is avoided; likewise you can switch from bridged to routed or just break one of the connections so it doesn't form a loop.

Ah.... *spanning tree* of course - I ran across that earlier on in my exploration of the NVRAM variables and forgot about it later on when I needed it.
Thanks!

Hi,

I like to do an suggestion to add a new feature to OpenWRT that is called Point-to-Point Bridge setup.

You see this mode only on more expensive Access Point's.

This is mainly used for: Backhaul / wisp / extend network / Etc...

Here is what i want to do with it:

Modem/wan <--> Main Router/AP <-- Network home --> AP-Bridge <-----point-to-Point-----> AP-Bridge <--> 8-port switch <--> Network

I do NOT want to use WDS for this kind of setup!

In the past I have to use two WAP54G AP's for such setup but now I like to do this with wireless-N AP or Wireless Router.
The antenna's have to be Dual-Polarization to achive 150 / 300mb/s connection but also with only one antenna the speed wil be higher than wireless-G !!!!

Any information is welcome!?