Hello,

I am looking for people that have more then 25 boxes running in one city using MT?

Can you knindly post your experience here?

I have invested $6000 last 2 years in wireless and was unable to reach anything more then 1 hotspot server connected to a adsl line.

All 4 wds i have are not working from time to time. The wdsses only need to repeat the server signal and this works but not stable enough.

I encounter much troubles from time to time, things that worked months before are now crashing day to day, just like that.
How more clients i get how the more the server start crashing.

I want to note, V2.8 and before was totally not ready for WDS.
Since V2.9 came out WDS went much better, but still i have the most stupid problems that can be prevented via simple things that unfortuny still not are implented.

- A hotspot and dhcp watchdog for example because i encounter more and more troubles as the network grows.
- A centralised wds monitor to see if all wds still are running.
- A loadbalancer that can route based on GB's because i have dynamic ip adresses that do not work with static, policy or any other routing in MT.

The biggested "pure Mikrotik" deployment we've done contains 42 access points. All running 2.8.26 or 2.8.28 with Atheros5212 and Prism2.5 miniPCI radios and WDS. It works quite well.

Is it just me that I dont understand, but why do you use wds? It is cheaper, but it has so many problems...
I mean, right way to do things is to have one card in ap mode, and other card with another directional antenna for p2p link...and everything is working much better than in wds mode...
just my 2c....

I tend to agree a bit with the WDS approach. We went from a all-Cisco 340 network, in the proprietary "Cisco WDS". Other than lightning, we rarely had reliability problems. It wasn't the fastest thing in the world, but it was pretty good broadband for the year 2000. Now we use Mikrotik with dedicated backhauls, and one major issue is when you have backhaul failures, then you need to reroute, or have a secondary in place.

With all WDS, you have a mesh network of sorts - it just finds the next strongest signal and attaches to it. A self-healing network, that's cheaper and allows you greater coverage per dollar. We spend anywhere from $2000 to 3000 for a tri-sector array and backhaul. By taking those 5 AP's (2 backhaul, 3 AP's) and making them all WDS, you get to have 5 towers for the price of one. With the proper dual-radio approach, you wouldn't lose speed either - I'm trying to figure out a way to utilize OSPF routing with dual-radios (or triple, 2 5ghz radios, and 1 2.4 radio for clients) to make a meshed MT approach work.

DirectWireless, do you have a visio or any type of drawing of your setup?
I'm having a hard time visualizing what you are doing, but it sounds very much like what we want to deploy.

No, I haven't even designed the mesh system at all yet - just throwing some ideas together that maybe we could put together into a working OSPF routed mesh dual,tri-radio Mikrotik install script. I would use EOIP tunnels to make the Wifi-client radios connect on the same subnet, to allow intra-cell roaming. On the backhaul end, I would use NStreme with polling. This would almost certainly require RB532's for max performance, but I can test with lower end. If it works (and auto-configures, heals) I would be willing to share my configuration with the rest of the forum, although I know I'm treading well into unknown waters,

I can imagine the power of a meshing CPE - not to mention if Ubiquiti comes out with 900mhz cards, a 900mhz meshing system - how powerful! Dual SR9's all mounted in a nice enclosure.

I'm almost 100% sure MT can do it, I do believe I have at least 3 dual-CM9 radio WRAP setups in my stock, if not more. I really only need a gateway, 2 repeaters. Then I can test to see if failover routing will work. I hope the concept of an automatic shutoff AP will solve one main issue - dead mesh loops. It wouldn't technically be a full mesh that would allow multiple simultaneous uplink backhauls, but rather a tree structure that in the case of a full gateway failure, could re-mesh from a further away gateway on the other end of the mesh.

If all went well, the affected network outages would shut down their AP's automatically and search for live AP's, and then reactivate the on-board AP when the link is stabilized back to the gateway. Creative scripting could allow regular bandwidth tests to determine if a link is stable, and change channels if necessary (on the AP side) and disconnect bad links on the station side.

No, I haven't even designed the mesh system at all yet - just throwing some ideas together that maybe we could put together into a working OSPF routed mesh dual,tri-radio Mikrotik install script. I would use EOIP tunnels to make the Wifi-client radios connect on the same subnet, to allow intra-cell roaming. On the backhaul end, I would use NStreme with polling. This would almost certainly require RB532's for max performance, but I can test with lower end. If it works (and auto-configures, heals) I would be willing to share my configuration with the rest of the forum, although I know I'm treading well into unknown waters

Why would you use polling on the backhaul? I thought polling was only useful in point to multipoint configurations.

Well if you use polling for the AP, don't you have to enable polling on the station side as well? I know polling on the AP side would be helpful, it would reduce a lot of hidden node issues and self-interference.

The biggested "pure Mikrotik" deployment we've done contains 42 access points. All running 2.8.26 or 2.8.28 with Atheros5212 and Prism2.5 miniPCI radios and WDS. It works quite well.

Are all 42 APs then sharing a single 11/54 megabits or is the capacity aggregated by multiple uplinks to the "wired" world?

The network is on a single Internet connection. They don't all share the same path to that gateway however.

The network is on a single Internet connection. They don't all share the same path to that gateway however.

Can you use only 5.5 megabits of upstream, or can you utilize more than that?

Being WDS, how is it routing packets through the network, avoiding loops - manual WDS entries, or STP?

Hammy - we're not limited to 5.5mb as both .11a and .11b are being used. I'm not sure what you are trying to get at...

DirectWireless - It's all static-wds right now. We'd like to come up with a more dynamic/redundant solution, like OSPF as others have mentioned. Not sure I'd want to deal with dynamic WDS and STP.

So you have network of WDS repeater omnis, or is it multiple radios with WDS? Like my network, I have set backhauls, with directional antennas, all set to WDS mode (point to point) for bridging.

Ultimately, given perfect LOS and tower proximity (<5 miles) I'd like to use 5ghz omnis for WDS without pointed backhauls -that way I have many possible redundant paths - most would be omni (or 120 degree in the case of towers that are along the ocean here (no need to send signal out the wrong way) but with a main AP, the rest wouldn't actually be repeaters unless there was a failure... I'd still rebroadcast with 2.4's, but instead of needing 2 redundant paths to a tower, it could simply use the strongest AP. Climbing towers and pointing antennas for us is a real headache, since we have to contract out to do it.

My only concern is that if a path is like this:

AP A --> AP B --> AP C

What happens when:
1. AP A sends a packet to a user on AP B's ethernet port - does AP C get it?
2. AP C sends a packet to AP B, does AP A get it?
3. AP A tries to transmit at the same time as AP C both to AP B - is it a collision?

Would NStreme with polling eliminate that issue? How fast would a 5ghz or 5ghz-turbo WDS setup be?

Happy New Year!

I'm looking at a whole town wifi cloud. WDS makes this much easier. However, I don't want to have the entire town limited to 5.5 mbit when a B user is signed on. I'd like to throw in several N-Streme backhauls to take care of that. That is, if WDS will let me do that.

*bump for myself as well as the others that have pending questions*

We use both directionals and omnis - you need to use what works. If you need distance and only have one possible link, a panel makes sense. If you can distribute and don't have a distance limitation (WRT gain and signal levels and noise) then an omni may make more sense. Every deployment is different, and there is never just one solution for them all.

I'd like to try out nstream with this type of setup but haven't yet. Creating "dumb" dynamic links based on strength can be problematic if there is interference or similar links as you might find your APs bouncing back and forth alot, which makes the latency do the same. Even with static WDS links, you'll see this if they are close enough (even though they never fully oconnect, they will try, sometimes a lot). This is why we went static.

DirectWireless, I think you had mentioned at some point in another thread about using OSPF instead of or along with WDS which would be very interesting to try. If I ever have any free time and a bunch of spare radios, I'll give it a shot Have you experimented with this at all?

However, I don't want to have the entire town limited to 5.5 mbit when a B user is signed on.

It sounds like you are interested in using b/g radios instead of just b?

I'd like to throw in several N-Streme backhauls to take care of that. That is, if WDS will let me do that.

There's no reason you can't do this, but perhaps you should separate your access points from your backhaul if your deployment allows for this as your throughput limitations would be at your APs instead of your backbone. How much bandwidth are you going to feed this network? How much bandwidth is allocated for each user? If you only have, say 3Mb of total bandwidth and user queues of 1Mb/user, a limitation of 5.5Mb isn't such a problem...Not sure what you had in mind though.

*sorry for the late reply

Yes, looking to run b/g, maybe even lock out B radios. Yes, looking to do wifi over B/G and then some PtP or PtMP (depending on scenario) N-Stremes with A. In one area I'd be looking to install maybe 50 WDS AP nodes. I would permit a certain degree of client - client for those that request it (put them on another VirtualAP), so within the area I would plan for a rather high amount of bandwidth usage, but maybe only feed that system with 15 - 25 megs. Different queues for different levels of service from 256kbit on up to 10 megs. If both nodes are on the WDS, would it travel that or if it was fewer total hops to go WDS - N-Streme - WDS to another node?

well I did have this idea the other day regarding WDS. What about using using dual radio, one for a backhauled, NStreme WDS (no clients), and a secondary that is an AP for clients. I'm thinking 802.11a on the backhaul, and b/g on the AP. On the WDS side, you could force the radios to only associate at 36Mbps or higher, so you have a fast network.

I was also thinking that rather than bridging the WDS links, which is the typical use for WDS, you could assign IP addresses to the interfaces, and enable OSPF - that way you could then dynamically assign route costs based on signal strength and or associated speed and still have a single channel WDS based network. The secondary AP cards, would then be EOIP tunnelled back to the main network so you could have roaming and a single IP subnet of public ip's available. Wirth VirtualAP, you could use multiple tunnels back to the gateway(s).

Would this OSPF idea work?

I did some testing with WDS (non-bridged), and I was able to assign an IP to the interface and ping across it. It is recognized in the OSPF section as an interface, so I am in the process of setting up an OSPF test network. However, I did notice the dynamic WDS interface will have no way to have an assigned IP address, so I figure I can write a script to do something like this:

Every few seconds, check for a WDS with no assigned IP address. If one is found, it would assign it an IP automatically, on the basis of using /30 addressing (2 usable per WDS link, one for the local link, one for remote). I imagine that maybe the most complicated part of this whole thing - making a workable, 10.a.b.x auto-address scheme. I was thinking originally that the a.b could be the router "ID", and each class C would be split into /30 subnets, which would somehow work in a master/slave relationship once the dynamic WDS interfaces were created. Maybe the higher MAC address becomes the DHCP server, and the lower becomes the client? I'm not sure, a lot of details, but I'll keep everyone posted.

directwireless, I am interested in roaming between APs.
can you tell us more about your tests and results ?