Anyone have a working “rr bonding” configs using wds that they can post?
Please also post details of what ports exactly you are putting into your bridge (ie…ethernet, wlan, wds, bonding interfaces…).
We have an ongoing problem getting this working properly (subnet-wide network issues caused by it including packet loss and latency…), and have not gotten very far with MT support, whom basically gave a partial example and stated that it worked fine in their labs. We however have problems every time bonding interface is turned on, and have tried every possible confi we can think of the past 7 days (see my other recent post for details…).
Waaaay behind on install deadlines now
Don’t think this is hardware related, as same hardware testing dual-nstreme seems to work fine. But we want the higher capacity in half-duplex mode, rather than the lower full-duplex capacity dual-nstreme offers + the link redundancy of bonding.
burstnet, we gave up on this. We have never been able to run rr-bonding in the same situation like you have - on some links, I prefer to have some “half-duplex” traffic faster in one direction than the other one so nstreme2 was not solution for our problem. We spent too much time trying to make this work before we gave up - should we have invested this time and effort to activities generating income, we would have our own fiber link laid down at this time.
It’s weird, cause I had it working perfectly/phenominally in the lab between two computers and two MT routers in-between doing the bonding over wireless bridge interfaces, BUT, as soon as I plugged the one end into our live network—it started causing the network issues.
Is that the same problem you had? packet loss / timeouts and latency?
Have either of you tried a managed switch that does link aggregation in some form or other? We did some quick and dirty trials with wireless links that seemed promising but then ran out of time.
However, we did get RSTP running last night on a licensed wireless DS-3 primary with a Redline backup link. Worked ultra slick. When running a MT bandwidth test you couldn’t see any loss when pulling the DS-3 plug. The test just slowed down from 45 to 20Mbits. That was using a pair of Netgear FSM726S which are around four hundred bucks each.
My point being that possibly a relatively “dumb” switch might be an answer to what you are trying to achieve…
There are days when I think MT has jammed too many features into RouterOS.
That is my backup/next plan, but would require multiple Mikrotik machines to accomplish (which may be better for overall hardware redundancy). It’s less expensive to do the link aggregation (bonding) with dual cards in two Mikrotik machines, rather than a switch requiring four Mikrotik machines instead. Just a shame, cause they advertise bonding, and it just does not seem to work properly…atleast from our time we put into it.
I’m hoping it is just a setting that we are missing and their support has not noticed. With the lack of documentation though, it’s hard to pinpoint. We tried the other bonding modes with no success either..aleast the other one or two that offered link aggregation and redundancy.
I tried bonding with MT about a year ago. That was obviously a lot of revs ago in 2.9… We could never get it stable which is why we moved to trials using a switch. I think that may be a better route.
In our case, we wanted bonding across a couple of ADSL links back to another MT on our core Internet fibre. We were going to use this as a failover solution.
But something else more important came up and we never got to complete the project.
<< BurstNET, I assume, you have sent support output files to support with configuration that doesn’t work for you. >>
Yes, I have…
They sent back a very basic config, which they stated they tested in the lab and it worked for them.
Well, I already had this working in my lab too, and it worked great between two computers. It is when it is put on a live network that it just starts to cause all kinds of problems—what appears to be same kind of affect as a network loop (it’s not as ips are different and no duplicate paths, etc…) or garbage it sends upstream. Same exact IPs and hardware in ap - > station or dual-nstreme works fine btw…it’s just the bonding that displays such issues.
Tried this with link agregation (802.3ad) over switches, going over two MT pairs, this weekend…and had issues as well.
Going to test direct from switch to switch this morning to confirm it works, and then I’ll know whether the MT units in between are causing the issues.
Bonding 2 wireless links (like transparent Nstreme-dual) is not a good idea. The bonding mechanism works with certain coefficient of productivity, and this coefficient is far from 100%.
Assume you have 2 CM9 cards in each router, one pair is working for upload, other for download. Assume that each link can give 50 Mbps (just example, my own record for 5G-turbo is 75 Mbps) of data transmission, so you are getting 50/50 Mbps non-aggregated (i.e. download speed will not exceed 50 Mbps even if there is no upload at all). You want to aggregate this 50+50 into good 100 Mbps and give a customer 90 Mbps down and 10 Mbps up.
But, the bonded interface has an effectivity of roughly 70%. Bonding these 2 interfaces and testing traffic in one direction will give you only 70 Mbps of aggregated traffic, not 100. If you test both directions, it is 35/35 Mbps. So if you configure assymetric link, you will have 60 Mbps download and 10 Mbps upload. Under normal conditions, it will be 50 Mbps down and 50 Mbps up. Let’s say - what you win in download increment, you lose twice as much in upload decrement.
Is it worth trying - it’s up to you. My point is - no. Wireless bonding is hard to set up since when you group interfaces into bond, MAC of the opposite WDS endpoint changes and you have to reconfigure WDS interfaces again. You need to have both routers on the desktop so you can configure them since you need to have console access to them (impossible to set it remotely on a working link). At least it was so in 2.9.10..14 (do not remember)
The bonding mechanism works with certain coefficient of
productivity, and this coefficient is far from 100%.
But, the bonded interface has an effectivity of roughly 70%
this is not true. Three 100Mbit cards bonded in Linux show around 285Mbit real throughput, while two bonded gigabit cards show >1800Mbps throughput. Gigabit interfaces face limitation of pci-x bus, cpu power, quality of tuning and software effectivity etc.
I would definitely like to see two wireless interfaces bonded. It is no use for us anymore as we had to have solution fast, I just would like to see it. This reminds me something like “vrrp is working” while several people here were reporting it’s not, and finally after ten versions, there were some changes which allowed to work it properly. I feel this is the same - nobody uses wireless bonding in real world, it could (or it should work) but it’s not.
