So the situation is like this. I have two RB951G connected over around 50m of FTP cable. The link is established as 1Gbps Ethernet.
Starting a bandwidth test between them, over UDP I've got around 250 Mbps, over TCP I've got around 120-130 Mbps.
I guess this is the maximum capacity of the routerboards. I can't seem to find what is the expected throughput for RB951G though....
What I'm interested in is, whether it's possible to terminate this line with Routerboards that are able to utilize the full capacity of the link? RB2011 perhaps... but I think RB2011 is essentially the same like RB951G with one additional switch chip.
Is there a SOHO Routerboard that is able to handle 1Gbps line rate, or close to that? Even if it is more expensive, it would still worth it....

What devices are you using to generate the traffic?

What devices are you using to generate the traffic?

I use the BTest Server and Client on both Routerboards... I guess running the test on the boards themselves influences the results...

Exactly. Connect computers that are able to generate the traffic behind and then measure the throughput.

Exactly. Connect computers that are able to generate the traffic behind and then measure the throughput.

But still... Are the RB951 boxes able to handle 1Gbps traffic. I really can't imagine this to be possible. Not that I complain about it though..
I'll test it with machines behind the routers....

It has to be G version and you have to set switch mode. Maybe in bridge with fastpath also. Of course it will not route/nat 1gbit.

It has to be G version and you have to set switch mode. Maybe in bridge with fastpath also. Of course it will not route/nat 1gbit.

I finally got the chance to test it. So, two RB951G, two hosts connected on the Ethernet ports - the speed that I was able to achieve was around 250Mbps ...

What was limiting the speed?

What was limiting the speed?

In general there was no limiting factor, so I'm a bit puzzled. I didn't expect though to get 100% utilization, maybe 60-70%...

As jarda remarked earlier, that shouldn't happen if the ports are all switched.

I tested 951G with 411GL and with 2*2 MIMO that was capable of 200 Mbit/sec tcp speed with Ubuntu Proftp server.
CPU was around 60%. I think this is a very nice performance.

I tested 951G with 411GL and with 2*2 MIMO that was capable of 200 Mbit/sec tcp speed with Ubuntu Proftp server.
CPU was around 60%. I think this is a very nice performance.

If this is a speed you get over WiFi - it's more than nice.
In my case though, I'm mainly concerned with the speed I could potentially get over the switched ports... And it is also around 200 Mbps, which is a bit lower that my expectations for L2 switching.

951G and 411GL are powerful devices, so that's why i achieved 200 Mbit/sec.
I do not remember exactly 951G's switch performance, but that was around 300 Mbit/sec with Proftp.
I have never tested with internal bandwith test.

I didn't expect though to get 100% utilization, maybe 60-70%...

If you got such CPU utilization, you clearly didn't put the ports in L2 switch mode.
Under /interfaces make one of the ports slave to the other. Then you will have pure L2 switching, with 0% CPU usage for that task. And you will get wirespeed switching.

Well, actually I did that - and this is the default configuration:
[bsotirov@router82.sotirov-bg.net] > interface ethernet export
# jan/08/2015 16:48:08 by RouterOS 6.24
# software id = 2DWW-HXRL
#
/interface ethernet
set [ find default-name=ether1 ] comment="Outbound Interface" name=ether1-gateway
set [ find default-name=ether2 ] comment=LAN name=ether2-master-local
set [ find default-name=ether3 ] master-port=ether2-master-local name=ether3-slave-local
set [ find default-name=ether4 ] master-port=ether2-master-local name=ether4-slave-local
set [ find default-name=ether5 ] master-port=ether2-master-local name=ether5-slave-local

According to my understanding - all of the "slave" ports are contralled by the switch chip. Or maybe not?

Yes, those slave ports as defined in your example are controlled by the switch chip and will not use CPU resources..

I you use winbox/webfig, take note that ports will be marked as "slave" even if they are connected to a bridge, so don't take a "S" flag as an indication of them belonging to a switch. Check it to confirm they have a master port...

Ports in switch should not be able to connect to the bridge and vice versa.

You are right. Trying to do both assignments will give an error.
I just wanted to point out, that seeing an "S" in front of a port doesn't necessary mean it is assigned to a switch group.

But it is not so long time ago when it was possible:
http://forum.mikrotik.com/viewtopic.php ... 63#p426191

It was corrected by mikrotik according to my report.

I'm a bit confused now... Are my ports controlled by the switch chip or not? Basically I've put the master port in a bridge - in order to have an IP address there and to be able to have L2 connectivity with the Wireless port...
interface ethernet switch port print
Flags: I - invalid
# NAME SWITCH VLAN-MODE VLAN-HEADER DEFAULT-VLAN-ID
0 ether1-slave-local switch1 disabled leave-as-is auto
1 ether2-master-local switch1 disabled leave-as-is auto
2 ether3-slave-local switch1 disabled leave-as-is auto
3 ether4-slave-local switch1 disabled leave-as-is auto
4 ether5-gateway switch1 disabled leave-as-is auto
5 switch1-cpu switch1 disabled leave-as-is auto

As long as you have a master port set for your ethernet port, like you showed:
set [ find default-name=ether3 ] master-port=ether2-master-local name=ether3-slave-local
then YES, they are in a switch.

And it is OK to have the master port put in a bridge together with the wireless interface.

I have to give it another try than... If I get 50% utilization of the link, it would be great.

mulțumesc

I just crashed in on the switch discussion. But to come back to the speed topic:
On NAT, you hit the CPU limit at about 100-150Mbps, depending also on the rest of the configuration (queues, bridges).
But for ethernet connections, these devices are really capable of wire speed.
I have a RB1100 which uses the same switch chip as the 951G, and it does full wire speed switching.
Not sure how the bridge with the WiFi interface on it affects these speeds, but the Atheros 8327 is a capable little piece of hardware.