this can be bufferbloat, or rb433 series problem…
One day later: This was not the cable. I changed the POE splitter and power supply. (Not sure which of both was the source).
Leo.
Anyone ever figure out a good explanation for this? I’m getting odd tcp performance issues. I’m not using any mikrotik wireless, they are strictly routers. Udp and tcp with multiple connections = full speed. Tcp with one connections t uns anout an 8th of the speed but only one direction.
We are also seeing issues with single stream performance with tcp. Multiple connections and udp full speed single connection tcp testing a small fraction of what it should be. Have not identified problem yet…
A post with almost one year duration and no opinion of mikrotik support? My clients are going to other ISPs, and I’ll have to go to another provider routers?
If you have a specific problem I suggest that you describe it in detail in a new thread. This thread contains references to a number of different problems and the post 2 was confirmed fixed by replacing (non-Mikrotik) switch.
@hapi
have youe tried to sihft the frequency on the wireless link 5-10MHZ on up or down?
Do the test from your big xeon machine TCP 1 send to the Omnitik and post.
We had a simmilar link 10km on 5500Mhz NV2 2ms. Good RX/TX -67 both chains 98-100% CCQ
At 5500 TCP 1 connection was only arround 20Mbit/s - I twas not enough for us, so we tied to shift the frequency down and up.
The best Connection was on 5505Mhz TCP raisd to 65mBit/s (CPU maxed out)
let us know if it helps -
Hi all,
we’ve had similar problems and it turned out to be a bufferbloat problem on one of our licensed links.
Our Setup looked like this:
[Server] <----GigE----> [C2960G] <----GigE----> [Licensed_150Mbps_Link] <----GigE----> [C2960G] <----FastE----> [Ubnt_RocketM5] <----300/300 Mbps----> [Ubnt_NB5M] <----FastE----> [PC]
The 150Mbps-Link was connected over GigE to the Switch, but since it was not able to transfer data at GigE wirespeed, it started to buffer as much data as it could => bufferbloat. When we switched the Interface of the Server to 100Mbps, our Problems went away, because the 150MBps-Link was able to transmit at FastE wirespeed so there was nothing to buffer.
The solution was to enable Flowcontrol on the 150Mbps-Link and the C2960G.
I don’t know if this applies to your Setup but maybe it helps to find the solution.
\
- Mat
My problem is the same of this topic subject: TCP performance
RB433AH ↔ switch 3com gigabit ↔ Rocket M5 – WIRELESS – Rocket M5 ↔ switch 3com gigabit ↔ RB433AH
Btest
protocol: UDP
bandwidth: 60Mbps
Protocol: TCP
bandwidth: 12Mbps
Similar results testing between PC stations at the ends, using the FTP protocol, in both directions.
People often expect that UDP and TCP performance should be the same but TCP performance is affected by factors such as latency and half duplex links so the fact that TCP and UDP performance differs doesn’t automatically mean that there is a problem in RouterOS.
Please, something new.
Please, do not even try to use TCP Btest on non-x86 hardware. It is misleading! 
What speeds do you get when testing directly connected to the Rocket M5 AP’s? What is the latency on the link?
Hi!
As informed, also tested at two x86 in the ends, FTP protocol, same result.
Network Speed Test:
66.39Mbps
ps.: The ‘Network Speed Test’ from the ubnt is UDP protocol.
I have not tested via FTP directly connected to the radio from x86, will perform this test.
Five Packet Count:
2.46 ms 64
2.73 ms 64
2.86 ms 64
2.61 ms 64
3.5 ms 64
I have five locations with the same equipment in the arrangement I mentioned, all with the same behavior.
I’m thinking of removing the switch and make new tests.
ps1.: the rb are running pppoe server
ps2.: sorry for my bad english.
Sarcasm doesn’t solve such problems.
Quote the “people”, and not me.
I see this effect, too.
Sometimes it is due to some packet loss. A single TCP-streams shows this problems much earlier than a udp connection.
Sometimes it stays unknown. E.g. I have a RB433AH on one side of a wireless link. Performance on the link is fine (bw test from the RB433AH to the other side), performance to the RB433AH from the ethernet side is fine.
Connections through the RB433AH over the link shows slow tcp-speed. Looks like a forwarding problem within the RB.
What is at the other end if the radio link in this case and how is the link set in terms of protocol, duplex etc?
Rb411ah nv2 ap/station routed r52n cards.
Hello all.
What is the best method for identify the real throughput of a path?
The real traffic in the network usually are TCP not UDP.
Infact the your problem is the difference of results between TCP and UDP tests.
There is no “real throughput”. If you do torrent UDP-Speed is what you get. If you do ftp you use a single tcp-connection for data. If you use a browser it depends on how much connection this specific browser uses. Some do prefetches.
speedtest uses different amount of tcp-connections depending on what performance it gets.