I have some PTP multihop links that doesnt perform well. Seems the bandwith drastically decrease while passing an hop.
The main problem is that if I test 2 point, the throughput is max. If i go throgh 3 hops, it will decrease (like half less each hop).
The things are worst with TCP than UDP.
For example, consider this scenario:
From RB-OmniTik to RB2011 i have like 40Mbps. From RB-OmniTik to the first RB 711G-5Hnd v.5.23 I have 7Mbps. Please pay attention that there are no new wireless hop, because the 711G-5Hnd is connected by wire to the RB2011.
BUT if i test the ethernet wired from RB 711G-5Hnd to the RB2011 it performs 98Mbps (!). So there is no a wire issue.
If i test from the RB-OmniTik to the last RB1100H2 i have 7Mbps.
I have tried a lot of configuration and a lot of example found on the forum, but them didn’t help:
upgranding and downgrading ROS versions, I also tested the new 6.7. No difference.
swap the hardware and use the same hardware for PTP
swap frequency (there is no noise/interference)
force the GigE ethernet to 100Mbps without auto-negotiation.
the nework work using OSPF and MPLS routing. I also tried to put all in bridge, and only L3 routing without MPLS. Same issue.
I tried to swap the RB2011 with RB750, RB450 and RB1100H2, same issue.
So, i think there is definitelly a bug in ROS in multihop routing or L2 frame forwarding, and the thing are worst where there are nv2/wireless hops.
A REPLY FROM MIKROTIK AND A SERIOUS DEBUG IS NEEDED. Otherwise, I will start to swap my hardware to a new vendor.
tried dual chain, single chain (selecting various combination of ht-chain0 ht-chain1). I think to a bug because the problem when passing the wired section as I decribed. But the wired ethernet segment works if you test it from the RB711 ethernet to the RB2011 ethernet.
The single PTP works. They doesn’t works and the thrughput decrease if you traverse more than 1 hop.
why this uge difference from A to E ? the only diffierence is a 2 meter long ethernet cable, that WORKS because D to E is 98Mbps. I expect A to E at least 40Mbps like A to D.
How do you test this link? Using btest or iperf? Remember that these mikrotik devices don’t have a very fast CPU and are not able to saturate most links when testing TCP performance using btest.
i have had these issues too and mikrotik cannot drastic explain loss on multihops.
infact bandwidth test in udp stops working in such scenario.
have had similar ROS on all devices.
gringoZ is probably right. Low throughput is often the result of high latency.
Lowering the latency of the link or increasing the TCP window size on the endpoints will probably fix your problems.
What is your speed when testing UDP?
We’ve seen this behavior, too. If latency is the problem you should not see it with modern Systems
like windows 8. There are built in mechnisms to increment the tcp window size automaticaly.
We think it has to do with queueing within ROS or may be the lack of ethernet flow control/pause frames.
If i’m right,i cannot adjust the window size on an edge router, using the mangle. I think that it is adjustable only by the client or the server, right?
Yes. The tcp “window” is at the endpoints of tcp communications. It is the size of all packets which can be transmitted until the Sender stops sending waiting for acknowledgement. It adapts automaticaly on modern systems (Win XP did not). Higher latency needs a bigger window size.
