Strange 802.11n with MPLS issue

RouterOS Wireless Networking
1

I have a point to point setup that extends across 2 miles. The radios are at -60 dBm signal strength. The sync rate is 117 mbps. CCQ is 95 to 98% both ways. When I do throughput testing using Mtik Bandwidth utility with UDP I can get 80 to 90 Mbps either direction. When I change it to TCP I can only get 20 Mbps maximum either way. I can further do bandwidth testing using speedtest.net and get the same 20 Mbps.

I followed the instructions here to get the MPLS transparent bridge setup:
http://wiki.mikrotik.com/wiki/Transparently_Bridge_two_Networks_using_MPLS

I have searched the forums and can’t find anything that describes using MPLS and this type of issue. If anyone has any ideas, I would be very appriciative.

Thanks,

-z

2

Well UDP is a connectionless protocol, just fire and forget so it will always be faster than TCP. You need to test the bandwidth THROUGH the routers, not from router A to router B…

computer A <-----> router A <-----> router B <-----> computer B

3

Here is the setup and the speeds I get:

Computer A <----> Mikrotik A (802.11n Radio) <-MPLS Bridge-> Mikrotik A (802.11n Radio) <-----> Cisco Router <------> Internet

Computer A is where I am testing from and used winbox with UDP to get 80 to 90 Mbps.
I then tested using iperf from computer A to a computer connected to my Cisco Router and got the same results with UDP and 20 Mbps on TCP transfers.
I then test using speedtest.net and am able to get only 20 Mbps download and upload test. These test use TCP also.

-z

4

what’s your package from your upstream provider???
it’s 100Mbps???

5

My upstream provides me an unlimited upstream, but I pay for 100 Megs. I have a Gig uplink to them and when I am plugged directly into my core switch I can get 100 Megs on speed tests.

The tests I provided earlier were not tested outside of my network. I was doing jperf, Mikrotik Bandwidth Test, and other local bandwidth tests to prove the mysterious 20 Meg TCP cap on my MPLS link.

Has anyone else setup an 802.11n MPLS wireless link and not had this issue?

Thanks,

-z

6

What is the latency of the computer you are testing to the bandwidth server? Maybe it’s a TCP congestion algorithm problem.

7

1 - 3 ms to the bandwidth test server. I have smokeping running on my network to monitor latency from the computer I am testing from.

-z

8

according to you latency, your link suppose can burst up again…

9

I have another pair of radios that I am going to setup this evening with the exact same settings to see if I can duplicate the problem.

-z

10

I haven’t had the chance to test the 2 other radios yet, been busy busy. But I did spend a day and try to move it over to WDS from MPLS and the quality sucks with WDS. I was getting 5 to 6 Mbps transfer. I did notice that when I do a data transfer or tax the connection, my CCQ drops to 30 to 40%. I switched back to MPLS and am able to get 25 to 30 Megs now. I also re-aligned the antennas without difference. I also tried to drop the radio output (dBm). I tried with single chains to try and isolate the issue to horizontal or vertical pol. Both signals were pretty good. I am stumped. The only other thing I am going to try is try repositioning the unit one more time and if that doesn’t work, then I am going to replace the Mtik with some Ubnt Rocket M5.

-z

11

Have you tryed to change freq. ? I have same problem but I see if I change frequency my speedtest up to 53 Mbps. My test is with frequency 5210 ( I have use “only for test” country japan3, superchannel for few minutes ).

12

i never try Ubnt Rocket M5 cause it have no much feature.
but if you look for transparent bridge only, i think you may share with us the result.