Could please someone who has successfully bound wireless interfaces, send me the config?
I've been trying hard, but i just can't get it working.
Btest from one mt to the other shows really nice numbers, but real throughput from one computer behind mt to another, behind the other mt, gets worse with each interface added to bonding.

My config is like this: i create eoip tunnel over an interface(i think i've tried all of them, wds, bridge+station, with and without nstreme, dual nstreme) and then bind them and bridge bonding to ethernet interface.

See this link, and if you succeed give us a reply!!!
http://wiki.mikrotik.com/wiki/Bonding_Examples

Well, that example wont work without some tweaking, wlan interfaces are on different networks (10.1.1.1/24) (10.0.1.1/24) and there is no ip connection, therefore eoip wont work

Hmm, one more thing, udp packets go fast, its tcp packets that get so slow, any clue why or how to fix it?

is there a problem with bridging in 2.9.50 version or something? i can get easily 60mb/s duplex between two mikrotiks just by routing traffic to one card and using another as reciever, but when i bridge ether1 with eoip, the traffic over the link is cut down to some 20mb/s.
Its same way if i use nstreme dual.

your hardware and signal strenght must be strong. What is your hardware and your signal?
We tried in the Lab. RB500 maximum TCP is around 20Mbps, RB300 40Mbps, RB600 40Mbps with maximum signal.

your hardware and signal strenght must be strong. What is your hardware and your signal?
We tried in the Lab. RB500 maximum TCP is around 20Mbps, RB300 40Mbps, RB600 40Mbps with maximum signal.

(Just agreeing with above quote and put to rest the minds of fellow noobs.)

It's true what they say - bigger is better! Or in this case, faster is better. Regardless of signal strength, the hardware needs to be able to handle the volume of data too. We tried an rb600; maximum internal speed was about 60 Mbps and maximum wireless link speed was about 41 Mbps over ~6km.

Picture a brand new highway as optimal signal strength; your horse and buggy will only be able to go so fast!

your hardware and signal strenght must be strong. What is your hardware and your signal?
We tried in the Lab. RB500 maximum TCP is around 20Mbps, RB300 40Mbps, RB600 40Mbps with maximum signal.

(Just agreeing with above quote and put to rest the minds of fellow noobs.)

It's true what they say - bigger is better! Or in this case, faster is better. Regardless of signal strength, the hardware needs to be able to handle the volume of data too. We tried an rb600; maximum internal speed was about 60 Mbps and maximum wireless link speed was about 41 Mbps over ~6km.

Picture a brand new highway as optimal signal strength; your horse and buggy will only be able to go so fast!

udp or tcp? our tests never got past 33mbps tcp. What setup were u using?

your hardware and signal strenght must be strong. What is your hardware and your signal?
We tried in the Lab. RB500 maximum TCP is around 20Mbps, RB300 40Mbps, RB600 40Mbps with maximum signal.

(Just agreeing with above quote and put to rest the minds of fellow noobs.)

It's true what they say - bigger is better! Or in this case, faster is better. Regardless of signal strength, the hardware needs to be able to handle the volume of data too. We tried an rb600; maximum internal speed was about 60 Mbps and maximum wireless link speed was about 41 Mbps over ~6km.

Picture a brand new highway as optimal signal strength; your horse and buggy will only be able to go so fast!

udp or tcp? our tests never got past 33mbps tcp. What setup were u using?

You never measure a wireless link speed in TCP. WiFi is Simplex and TCP is a full duplex protocol. I am doing +60 Mbit UDP on a Routerboard 333 one direction. If i add a radio/antenna i can either bundle 2 links and get around 80Mbit UDP or use it as 2 seperate links, one for each direction and get full duplex, ie + 60/60Mbit TCP

/Henrik

WiFi is Simplex and TCP is a full duplex protocol

Isn't WiFi actually Half-Duplex?

WiFi is Simplex and TCP is a full duplex protocol

Isn't WiFi actually Half-Duplex?

Half duplex / Simplex. This is why TCP goes slower. UDP is connectionless and TCP needs an ACK back for every packet it sends.

/Henrik

Just a clarification:
simplex >< half-duplex

What is the point in testing a link with UDP? Yes, we see higher numbers that way, but for most "real" traffic, TCP seems to give us much more accurate results of what the links will actually handle. I guess it's nice to see what it's capable of, but it just doesn't seem to be a very real world test.

So, I gues RB333 is the board to take for highest bandwidth (or of course x86 setup)..
Because when I've tried for example RB133c with R52, I only got max 25mbps when CPU hit 100%..

We reached 150Mbps UDP but TCP we never reach above 30Mbps!! We bonded 3 wireless cards.
Any body knows why?

TCP bandwidth test? What boards?
Do the bandwidth test through the boards and not directly between them.

Yes, TCP bandwidth test. The test done on two RB600 latest O.S.
Yes, TCP and UDP bandwidth tests done between the two boards directly!! Should we reach more? How much?

Yes, TCP bandwidth test. The test done on two RB600 latest O.S.
Yes, TCP and UDP bandwidth tests done between the two boards directly!! Should we reach more? How much?

You may experience that the TCP test is slower when you do it directly between the boards - we have seen it too.
You should use test TCP bandwidth test through those routers then you will get the best speed
For UDP you can run directly between the routers, but I also recommend to do it through the routers.

I absolutely can agree the bonding problems under 3.X.
Anyhow i have never reached a higher bonded troughput than a single link without bonding.
I'm using RB333 with multiple R52H Cards and directional Antennas over approx 11.1km as a PTP link.
When using all links together, each as it's own bridge connected through vlans, each Link is cappable of transfering approx 11.1Mbit TCP send/receive Traffic passing it. So RB333 with these three cards is handling total combined approx 66.6 Mbit (3 x 11.1Mbit Send, 3 x 11.1Mbit Receive) without any problems at the SAME time.
But when combining these three links into one BONDING_RR, the performance degrades below a single link with average TCP speeds from 4 to 12 Mbit in one direction.

The BTEST Server and Client are located on each end of the link on P4-Based PC's and thus don't modify the Testparameters on the RB333's.

I have no testing possibility to do comparements with 2.9.x on Bonded Links over wireless.

Does somebody succeed with Bonding over EOIP over Wireless with 3.x oder 2.9.x ?

Kind Regards,
Wolfgang

Try to fix the rate to 24Mbps "try other rate also" and don't keep it auto.
In lap. we found that if signal strenght is deffirent between each two cards, which is the usual case, M.T O.S try to send the best rate accourding to signal strength and here you will notice the drop in bandwidth. If you force your rate in a certain choise you will reach better bandwidth.
Try it in office first.

All Bonded cards have fixed equal bandwidth. Target speed which works well allone is 36Mbit each.
I tried to lower bandwith down to 18mbit symmetric on all cards at the same time with the reproducable results, that bonding is no increase in overall speed.
As i already told before, all cards together doing high traffic very well with 36mbit bandwidth each.

Yes, TCP bandwidth test. The test done on two RB600 latest O.S.
Yes, TCP and UDP bandwidth tests done between the two boards directly!! Should we reach more? How much?

You may experience that the TCP test is slower when you do it directly between the boards - we have seen it too.
You should use test TCP bandwidth test through those routers then you will get the best speed
For UDP you can run directly between the routers, but I also recommend to do it through the routers.

This was not a solution, only a "How to test speed on links".
Problem(all tested with tcp ftp traffic ~1400 avg packet size -- nearly ideal ):
rb600 - rb600 wds turbo: 4.21MByte/s (same with rb333)
rb600 - rb600 wds turbo nstreme: 6.21 - 8.1 Mbyte/s (same with rb333)
rb600 - rb600 nstreme 2 : 5Mbyte/s up/3 Mbyte/s down (better with rb333 but this is installation specific)

bonding:
bonding over wireless only works if you dont enable nstreme (speed nearly 7Mbyte/sec same as nstreme link).
bonding over 2 eoip interfaces over 2 wireless nstreme can reach 35Mbit/s on each device (70 Mbit/sec)
(bonding method balance-rr this works other methods can't works with scenario above)

cpu max at ~70%

conclusion: rb600 more powerful than rb333 but you unable to use this power.
This is the problem.

Yes, TCP bandwidth test. The test done on two RB600 latest O.S.
Yes, TCP and UDP bandwidth tests done between the two boards directly!! Should we reach more? How much?

You may experience that the TCP test is slower when you do it directly between the boards - we have seen it too.
You should use test TCP bandwidth test through those routers then you will get the best speed
For UDP you can run directly between the routers, but I also recommend to do it through the routers.

This was not a solution, only a "How to test speed on links".
Problem(all tested with tcp ftp traffic ~1400 avg packet size -- nearly ideal ):
rb600 - rb600 wds turbo: 4.21MByte/s (same with rb333)
rb600 - rb600 wds turbo nstreme: 6.21 - 8.1 Mbyte/s (same with rb333)
rb600 - rb600 nstreme 2 : 5Mbyte/s up/3 Mbyte/s down (better with rb333 but this is installation specific)

bonding:
bonding over wireless only works if you dont enable nstreme (speed nearly 7Mbyte/sec same as nstreme link).
bonding over 2 eoip interfaces over 2 wireless nstreme can reach 35Mbit/s on each device (70 Mbit/sec)
(bonding method balance-rr this works other methods can't works with scenario above)

cpu max at ~70%

conclusion: rb600 more powerful than rb333 but you unable to use this power.
This is the problem.

If you set up 2 WDS links with Nstreme in routed mode you can get 70/70Mbit/s FD. For some reason protocols like Dual-Nstreme, Bonding and EoIP doesn't work well on V3.

/Henrik

We succeed with wds bonding. Now we wanted to apply it with Nstreme enabled to get advantage of diabling CSMA.
Is it possible to do bridge to bridge "not wds-bridge" with WDS and Nstreme enabled?
Any one has an idea?

I made a test on two RB333 with two R52 each.

Setup:

Add a bridge to each router
Add ether1 to the bridge on each router
Add one ip (i.e. 192.168.10.71/24) to bridge1 on R1 and one ip (i.e. 192.168.10.72/24) to bridge1 on R2

R1: wlan1 and wlan2 in bridge mode, 5ghzTurbo, good channel separation, dynamic WDS, no default WDS bridge
R2: wlan1 and wlan2 in bridge mode, 5ghzTurbo, good channel separation, dynamic WDS, no default WDS bridge

Remember to set wlan1 on R1 on same ssid/freq as wlan1 on R2
Remember to set wlan2 on R1 on same ssid/freq as wlan2 on R2

When connected, add the corresponding mac-adresses in connect table on each router and disable default auth.

Copy the two dynamic wds-interfaces to static ones on both R1 and R2, and change wds-mode on all four wlan to static wds

Again on each router, add a bonding interface (balance rr), and add the two static wds interfaces to it as slaves. Add arp-link monitoring on each with the opposite ip

Finally, add the bonding interface to the bridge on each router and check connectivity. Everything should be fine - we now have a bonded wds-link in transparent bridge mode!


Now to my question: When I test the above setup with bandwidthtest using one PC in each end, the result is consistently that I get 40 Mbps udp in one direction and 27 Mbps in the other - and it doesen't matter if I run the test in one direction or in both. The interfaces are loaded equally and everything looks normally (54 Mbps*2 on both interfaces, approx -50 dB, S/N 40 dB etc.).

It is obvious, that more cpu-power will give you better performance, but this is not the explanation for the uneven load on the two wireless links ?????

Regards KimC

OK, I think the problem is bad nic's. Changing one nic on one pc gave the following with udp/both: 67Mbps/42 Mbps, and with udp one-way I get about 90 - 95 Mbps.

This is lab-results, however, but it is surely impressive !!!

Mikrotik: Why don't you document this stuff ? This is what everybody are asking for...

Regards KC

Hi Bill

Pls take a look at my previous post above. You should be able to get going with the posted info.

Regards Kim C

KimC, we need it with Nstreme to get benifit of disabling CSMA.
Have you succeed bonding with Nstreme??

I have applied the exact config as described by KimC above with two RB600 with R52H cards running ROS3.6. However if I copy a file between two devices on either side of the link the bonding performance with two interfaces is worse than when I use just one of the links in the bonding interface.

While I'm copying the file I disable WLAN1 and my performance improves (from 18Mb/s to 35Mb/s). If I then enable the WLAN1 again the performance decreases again. I then disable WLAN2 and again my performance improves again to similar numbers as mentioned previously.

Therefore my performance on each of the individual links is better than with them combined, i.e. bonded.

There is something wrong with the bonding performance on ROS!

You have to pay attention to your wireless connections. Not everything is about bonding - all the basics must be in order.

When you have fixed your links (if on your desktop, then dont use more antenna, than abt 2 cm of paperclip. Make sure, that the signal is around 50 dB on all radios), you will see, that it really works.

I'm afraid I'm not able to repeat your results. Both WLAN1 and WLAN2 on either side have wireless registration of 45dBm or better and connect at 54Mbps for both rx and tx. In fact I have even fixed the rates on either side.

CPU is not an issue < 20%. With both iterfaces enabled the best I can get by copying a file through the "link" is 24Mbs. With one of the interfaces didsabled I get up to 38.8Mbps.

WLAN cards are configured for 5GHz turbo mode, first available channel for WLAN1 and last available channel for WLAN2

As KimC says above, make sure your link is a -50 or so. Too strong of a signal causes problems as well. Imagine I'm yelling in your ear while your bud is talking at a normal level 5 feet away.

Same thing goes for the radios. Anything stronger than a -50 in my testing shows a loss of CCQ. Below -40 shows a signifigant drop in performance. My target for all of my CPE is -70, for my p2p is as close to -65 as possible.

configure how did you do?

Can you please me?

Best regards

Any progress in this issue? I'm experiencing exactly the same problems as nicopretorius. I've tried everything mentioned here with no luck.. Still single link in turbo mode performs better than two identical bonded.

I've noticed that I'm also experiencing a problem I don't quite understand. Let's say I have two working wireless links (I can see both of them in registration table) and when I add bonding interface on R2 (in which wireless interfaces work in station-wds mode), my wlan2 link disconnects and it tries to reconnect (as seen in R1 (bridge mode)) with MAC address of wlan1 radio in R2. I can fix this problem by changing WDS MAC in R1 wlan2 wds interface, but I don't think this is a good way of setting it right. I also believe this could by the reason of my poor performance. I don't think it's normal to have two WDS interfaces with same MAC belonging to different masters (wlan1 and wlan2).

So did anyone ever get Nstreme working with this type of link. I have the same thing and it works great but I want to use Nstreme or NV2.