Wonder if any of you guys can help me with an issue im having with Mikrotik Router boards its driving me crazy!

We normally use the RB450G as the edge router (PPPoE and queuing) and it has been working perfectly and on the bigger locations we have been using the RB800 and they work perfectly.

Obviously as the new RB850Gx2 has come out we thought it would be an ideal RB450G upgrade but this has caused some problems.

All Routers are on 6.24

Router A (Core) ------ Router B (Edge Router) ----- Router C (Access Point Switch)

Router A = RB1100AHx2
Router B = RB850Gx2 (PPPoE Server / Queuing) – Also the same with RB450G
Router C = RB450G – Just being used as a switch and BW tester

Obviously wireless links between each router. In the current setup if I run a bandwidth test between routers these are the results :

Router B to Router A = 95Mbits TCP on 1 Connection
Router C to Router B = 40Mbits TCP on 1 Connection

Router C to Router A = 15-16Mbits TCP on 1 Connection.

I have changed all the wireless links and cat5 cables and it’s still the same. I went to site today to swap out the RB850Gx2 to a RB800 and if by magic the problem has gone here is the results now :

Router B to Router A = 95Mbits TCP on 1 Connection
Router C to Router B = 75Mbits TCP on 1 Connection

Router C to Router A = 65Mbits TCP on 1 Connection.

I was convinced this was some kind of wireless problem or interference. I know there is still a slight bit of speed drop on the links but no way near like it was before. This is also a problem with the RB450G.

All the Edge Routers are connected to our core network with gigabit ethernet on the antennas.

This is a common problem not just at this site. Im banging my head against a brick wall all the time with the bandwidth to the client even on speedtest.net gets stuck at 16mbits.

It feels like it’s some kind of Ethernet buffer in MikroTik software. I have noticed the RB800 has 1600 L2MTU?

Any ideas would be grateful!

Stewart

Run the same tests with UDP and with many TCP connections, to find of it is TCP or hardware bottleneck.
In any case test from or to RB itself is not good idea for benchmarking as devices need to generate traffic and networking CPUs are not very good at that.