Hello.
I have searched, and have not found something that can be of use to me, in the following scenario:

I have three CRS125-24g.
CRS A is having internet access, from an uplink, on port 1. Separate domain server with dhcp server, on port 2.

I want to link CRS A with :
- two ports, 23 and 24 to CRS B, to ports 1 and 2
- two ports, 21 and 22 to CRS C, to ports 1 and 2

This will be a bridge, no spanning tree, nothing fancy.
How can i configure it so I don't have a loop between the two interfaces each of these routers is linked to each other, and so that the traffic is forwarded, "plain" (no routing, no vlan, no anything) between them ?

[ Two more questions:
a. will the link provide failover ? (eg. one cable goes out, the traffic goes only trough the other ? [EDIT: It does !]
b. will the link aggregate ? I could have more than 1 g traffic on the link between the two switches ? ] [EDIT: Does not seem to do it. even with balance-rr checked]

If I missed something on the forum or the wiki which can help me, direct me there. If not, please help me "on the subject".

Thank you in advance !

Somehow I managed to do it, by bonding the 2 interfaces, on each side, and bridging the bonding with the master interface set for the rest of the ports.
But on a btest I get 650 Mbit/s max.
That is all i can get from a crs ?

Btest uses the CPU to generate and measure the traffic, so I would certainly expect to get poor results with Btest on a CRS (or RB2011 or RB9xx series) unit. It's recommended to test between two devices that are both connected to the CRS.

Thank you.
I know that.
The test is done exactly like that.
From one CRS to another, passing trough a third CRS.
Still, could not get better result.
CPU 70%-80%. Never 100%. Plenty of lost packets.

If anybody is willing to give me another solution ?
Now the config is as this:
Ether 1 and ether2 - bonding1
ether 23 and ether24 - bonding2
ether4 to ether22 - slave, master is ether 3

Bondings go to other CRS, configured the same way, to form a large switched network. ( 80 pc's, but on different locations, so from there is the need for multiple switches)

bridge between bonding1, bonding2 and ether3
cpu: idle - 3-7%, on btest - 70-80%.
Traffic - 650Mb/s, or close to it, one direction, udp, 1500 size packets. Worse on tcp.

Got an eye here:http://routerboard.com/CRS125-24G-1S-RM
Still, faaaaaaaaar away.

ANY suggestion is welcome.

Using any CRS as your bandwidth test generator is going to limit the test, pure and simple.

The best way to test would be http://en.wikipedia.org/wiki/Iperf on two computers (which will easily have enough CPU to saturate a 2Gbps link).

But on a btest I get 650 Mbit/s max.
That is all i can get from a crs ?

The CRS125 has a 1Gbps link between CPU and switch. The btest will be running on the CPU, so the most you can hope for from a btest running on a CRS125 would be 1Gbps [or, ~940Mbps TCP].

It combines the best features of a fully functional router and a Layer 3 switch

Clearly, doing L3 at wire speed isn't one of "the best features"

The best way to test would be http://en.wikipedia.org/wiki/Iperf on two computers (which will easily have enough CPU to saturate a 2Gbps link).

The PCs running iperf would themselves need a 2G link to the switch in order to do this.