Hi all,

I would like to know what are the best practices when using bonding. I am running 8 RB532 Antennas (4 pairs) connected to with 2 x86 (one on each side) servers running an Routeros IDE Flash. All hardware is running RouterOS 3.1. I set up bonding and eoip-tunnels on the x86 servers to bond the 4 antenna links on one side and the other 4 on the other. All works fine but i am a little disappointed how the bonding is working.

Each wireless link manages 17Mbit/s but whenever i add another link with the bonding each link falls to 14Mbit.

If the third link is added, all 3 links fall to 12Mbit

if the fourth link is added , all 4 links fall to 9.5Mbit!

Bonding is creating too much overhead and i am losing almost 50% of my total bandwidth. I am bonding with balance rr since i want all 4 links to be active and have maximum redundancy (and eventually maximum bandwidth)

I am using static routes on each antenna and on the x86 servers to get to each other.

I am doing something wrong here?

cheers!

Been there, seen that.
i guess you use config like this:
lan-bridge-eoip tunnel - bonding - 4x(eoip- ip connection over wifi)
just forget it, bridge and bonding wont go together, use something like bgp nexthoops instead bonding

thanks for your response!

I am not really using 'bridging" since i am routing all over each antenna, (the rest if the configs you mentioned are correct) but i think the issue here is bonding! It works pretty well over 2 links, but once you increase the number ... no comment! Bgp would be a nice challenge.. never ran it on these boxes. Will consider it! 10x!

each bonded slave will affect the rest - not any one can go faster than the others... so if you have 1 link that's bad the others suffer. It's all about packets and their ordering - if they get out of order things start slowing down to compensate. Maybe the additional radio interference (?) is causing them to slow down ?

Each link alone goes up to approx 17Mbit, interference? well, at the moment they are pretty close to each other since they are in a lab environment before deployment. I configured the antennas using 5Ghz band, specific scan-lists added (no over lapping) As far as i know 5ghz should have no overlapping channels so that should reduce the possibility of having interference (correct me if i am wrong). Also all antennas are switched on at a go, so if there was interference it would have shown. I simply enable / disable the eoip-tunnel of the antenna link from the x86 and once the bonding detects the link is up, it recomputes and starts to forward again. Only that the individual bandwidth decreases at each added link!

It seems to be how the bonding balance rr protocol works.. The other bonding options worked well, but they are not exactly what we had in mind...

cheers guys!

are you able to send 17mbps across all 4 at the same time (non bonded)? Im wondering if sending 17mbps across one would affect the other 3 ... while when the other 3 are sitting idle you don't notice the interference. Basically if its idle the interference might not be noticable?

i see.. .. as soon as i find a slot today i will try it out and let you know!

Will keep you posted!

so....
i ran some bandwidth tests from Antenna to Antenna, all managed to achieve 17Mbit simultaneously. That means it is not an interference issue (nice try though!) I am actually looking into BGP now.. But i need a good loopback interface. Hope the bridge / loopback interface is stable. Will give this a shot and see how it goes.

Cheers!

good testing - that rules that out. I wonder if you went back to 2.9.50 if you'd have the same problems.

originally the x86 come with 2.9.50. I had the same issue hence why i upgraded to 3.1. The Antennas required to be upgraded anyway (shipped with 2.9.38) so i installed 3.1. From 10Mbit they inceased to 17Mbit. (impressive! obviously still have to test them in the open air) Installing 3.1 on the x86 i found that the speed was more stable so i am definitely liking 3.1 ! The issue with the bonding persisted. There must be a solution out there.. i just need time to find it!

Is that 17mbps half-duplex or full-duplex? If half, and on x86, you're way down on what you should be getting.

We have one outdoor RB333 link with 2x radios on each side, running Nstreme Dual, with 40mhz channels. We get consistent 40mbps+ full-duplex TCP, and 60mbps full-duplex UDP over the 1km link using two small panel antennas on each end.

half duplex TCP. I'm using a 500 series board. All antennas are loaded with security configs. (access-list, connection-lists, WPA2-PSK) which would slow the link down a little. There are also multiple ways of configuring the antennas. WDS bridges / AP or Station approach, 2/Nstreme, polling settings etc.
These all produce different results. Dual Nstream for example is not my case since i want full duplex on all antennas running 100% redundancy. I have to stick to the firm's requierments unfortunetly. (also antennas are not the latest model) The antennas at the moment are not aligned. In a lab i can't really align them as should, I am hoping that once aligned properly i achieve better results. For the moment i am focusing on the load balacing since bonding is not concinving me very much i also have to include (i did not mention it yet) another 2 x86 (slave) servers to run VRRP just in case one of the x86 fails! So what ever i do must also work well with VRRP. Anyway, first things first..

thanks dudes.