Page 1 of 1

Bonding two Dynadish 5 WiFi links

Posted: Sat Nov 25, 2017 3:17 pm
by bbahes
Hi!

I'm trying to speed up WiFi link for users on second site.
Current design is:
bonding-mikrotik-current.png
Since I have spare 2 x Dynadish and 2 x CRS I wanted to create bonding of these two links.
So desired design is:
bonding-mikrotik-desired.png
I have tested desired design but I get worse performance.

Can anyone give me suggestion or resolution where I'm doing wrong?

I have also included configuration export. Dynadish 5 links have same configuration and I have included only from second pair.

Many thanks!

Re: Bonding two Dynadish 5 WiFi links

Posted: Sat Nov 25, 2017 11:08 pm
by markmcn
As you have only posted 2 of the 4 config's from the DynaDishes I can't check this but have you confirmed you're not using the same frequency on both?
Also how close are the Dishes on both site, You might need to look at putting some space between them so you don't have an I.F interference between them.
I haven't used the Dynadish so I'm not sure about how good the screening is on them for close proximity use.

How is the performance on the system with a single link and the BOND interfaces still in use but with only one path?
One thing to remember that the CRS is a switch with a 600MHz cpu and that bonded interface traffic has to be processed by the CPU there is not
hardware processing of bond traffic. when testing the link what is the CPU at on the CRS devices?

I hope this helps.
All the best.
Mark

Re: Bonding two Dynadish 5 WiFi links

Posted: Sun Nov 26, 2017 7:42 am
by bbahes
As you have only posted 2 of the 4 config's from the DynaDishes I can't check this but have you confirmed you're not using the same frequency on both?
Also how close are the Dishes on both site, You might need to look at putting some space between them so you don't have an I.F interference between them.
I haven't used the Dynadish so I'm not sure about how good the screening is on them for close proximity use.

How is the performance on the system with a single link and the BOND interfaces still in use but with only one path?
One thing to remember that the CRS is a switch with a 600MHz cpu and that bonded interface traffic has to be processed by the CPU there is not
hardware processing of bond traffic. when testing the link what is the CPU at on the CRS devices?

I hope this helps.
All the best.
Mark
Here are configuration exports from second Dynadish 5 pair.
Both sites have same physical distance between antennas. It's 1m.

And here is CPU info:
CRS_CPU.PNG
I have tested link with BTest and ping:

Ping results:

from BON2 to BON1
sent=100 received=82 packet-loss=18% min-rtt=1ms avg-rtt=4ms max-rtt=20ms

from MT2 to MT1
sent=100 received=100 packet-loss=0% min-rtt=1ms avg-rtt=4ms max-rtt=14ms

and BTest results:
BTest.png
BTest was done at the same time between:
MT1 (server) < MT2 (client)
BON1 (server) < BON2 (client)

Re: Bonding two Dynadish 5 WiFi links

Posted: Tue Nov 28, 2017 10:32 am
by bbahes
markmcn wrote:
As you have only posted 2 of the 4 config's from the DynaDishes I can't check this but have you confirmed you're not using the same frequency on both?
Also how close are the Dishes on both site, You might need to look at putting some space between them so you don't have an I.F interference between them.
I haven't used the Dynadish so I'm not sure about how good the screening is on them for close proximity use.

How is the performance on the system with a single link and the BOND interfaces still in use but with only one path?
One thing to remember that the CRS is a switch with a 600MHz cpu and that bonded interface traffic has to be processed by the CPU there is not
hardware processing of bond traffic. when testing the link what is the CPU at on the CRS devices?

I hope this helps.
All the best.
Mark


So I did test again with CPU panel open and I noticed that when BTest started it pushed CPU load to 100% after that it drops connection and CPU load goes back to 1-3%.
So I can confirm that CPU is main problem here. We are going to buy new switch router with stronger CPU. Any advice? RouterBOARD 3011UiAS-RM ?

Re: Bonding two Dynadish 5 WiFi links

Posted: Tue Nov 28, 2017 12:41 pm
by td32
you must test final client(pc) to client(pc), testing is cpu hungry and running it on the same mikrotik device will not show real link results since cpu is used for generating traffic and the remaining for routing it.

Re: Bonding two Dynadish 5 WiFi links

Posted: Tue Nov 28, 2017 4:41 pm
by bbahes
you must test final client(pc) to client(pc), testing is cpu hungry and running it on the same mikrotik device will not show real link results since cpu is used for generating traffic and the remaining for routing it.
I did in first run. It was really slow. We have ordered new RouterBOARD 3011UiAS-RM and will try with them.

Re: Bonding two Dynadish 5 WiFi links

Posted: Wed Nov 29, 2017 4:02 pm
by markmcn
You have to remember that Bonding interfaces is also CPU bound

Re: Bonding two Dynadish 5 WiFi links

Posted: Thu Nov 30, 2017 3:25 pm
by bbahes
You have to remember that Bonding interfaces is also CPU bound

Even with new devices RB3011UiAS same problem.

Re: Bonding two Dynadish 5 WiFi links

Posted: Thu Nov 30, 2017 10:43 pm
by markmcn
Have you tried leaving the rb3011's and bond interfaces in place and only tried it with a single link in the bond?
This will tell you if it's an issue with the Dynadishes or the devices doing the bonding?
If things improve with one dynadish then we can look at that. If they don't then we need to look at the bonding.
How are you testing? Can you post some results showing actual performance figures?
As was pointed out earlier you need to test with PC's NOT the bandwidth testing tool.
Also what are the CCQ figures you're seeing on the links

Re: Bonding two Dynadish 5 WiFi links

Posted: Fri Dec 01, 2017 7:54 am
by bbahes
Have you tried leaving the rb3011's and bond interfaces in place and only tried it with a single link in the bond?
This will tell you if it's an issue with the Dynadishes or the devices doing the bonding?
If things improve with one dynadish then we can look at that. If they don't then we need to look at the bonding.
How are you testing? Can you post some results showing actual performance figures?
As was pointed out earlier you need to test with PC's NOT the bandwidth testing tool.
Also what are the CCQ figures you're seeing on the links

I did test on site with two ethernet cables only on bonding interfaces on CRS 109-8G-1S-2HnD-IN before going on with WiFi link. The test was OK.
bonding-mikrotik-test1.png
Testing is done only on Mikrotik devices using BTest feature and ping. Currently only single link is connected in bonding:
bonding-mikrotik-test2.png

CCQ is interesting:
bonding-mikrotik-CCQ.png
One thing I noticed as I observed CCQ value is that initially it was 70% and gradually ove one hour it went to 6%.
In that hour all users on remote site started using network.


I can test PC(site2)-Server(site1) today if that matters given the CCQ values?

Re: Bonding two Dynadish 5 WiFi links

Posted: Fri Dec 01, 2017 2:30 pm
by markmcn
Quick thoughts, I'd say you could turn down the TX power level's a bit to bring RX into the -60's.
Have you checked to ensure there is enough physical separation between the dynadishs?
Have you checked they are not on the same or very close channels (I.e Overlapping).
The poor CCQ is the problem. If you one run the link with one pair of dynadishes what is the CCQ like normally?
Assuming you have a clear path between sites it really sounds like the dishes are interfering with each other.
Assuming you are on different non overlapping channels and the CCQ only goes low when you use both pairs it sounds like you may need to add a bit more physical separation between the dishes.

Re: Bonding two Dynadish 5 WiFi links

Posted: Mon Dec 04, 2017 8:21 am
by bbahes
Quick thoughts, I'd say you could turn down the TX power level's a bit to bring RX into the -60's.
Have you checked to ensure there is enough physical separation between the dynadishs?
Have you checked they are not on the same or very close channels (I.e Overlapping).

The poor CCQ is the problem. If you one run the link with one pair of dynadishes what is the CCQ like normally?
Assuming you have a clear path between sites it really sounds like the dishes are interfering with each other.
Assuming you are on different non overlapping channels and the CCQ only goes low when you use both pairs it sounds like you may need to add a bit more physical separation between the dishes.
I have checked, distance between antennas is more than 1m.
As for channels. They are on 5825/20-Ceee/ac and 5145/20-eCee/ac.

As for CCQ I will have to check with company that set these two links up. I will report back.

Re: Bonding two Dynadish 5 WiFi links

Posted: Tue Dec 05, 2017 11:17 am
by markmcn
Well with a receive level in the -50's you need to turn back the tx a little, They are screaming at eachother .

Re: Bonding two Dynadish 5 WiFi links

Posted: Sat Dec 09, 2017 4:14 pm
by bbahes
Well with a receive level in the -50's you need to turn back the tx a little, They are screaming at eachother .
Which one? I'm still waiting for guys that installed this link to come and check this problem.
I see on all four Dynadishes Tx Power Mode: default.

Re: Bonding two Dynadish 5 WiFi links

Posted: Tue Dec 12, 2017 8:23 pm
by markmcn
to answer your question any unit with an RX in the -50's then you need to turn down the tx power of the one it's talking to,
Chance are you will need to turn down the tx power on them all

Re: Bonding two Dynadish 5 WiFi links

Posted: Wed Dec 13, 2017 7:06 am
by bbahes
to answer your question any unit with an RX in the -50's then you need to turn down the tx power of the one it's talking to,
Chance are you will need to turn down the tx power on them all
The company that installed two links tested them yesterday. We did turn down little bit power, but in the end problem with bonding was still there.
Then we went to secondary site and tested connection from clients to internet and we could clearly see that bonding is actually working as it should.
In first tests I did not test connectivity to internet but only to servers on primary site. Doing this test gave us clear picture bonding is working (we could see traffic being split on two ports).
The problem is that traffic to LAN on primary side is interrupted in some way that we need to find out how and why with company that controls DELL SonicWall UTM.

Thanks for all responses! I will report back if we get chance to check SonicWall logs.

Regards!