Hello
I have a CRS 125 and noticed that the routing speed is a miserable 30-40 Mbps. I obviously must be doing something wrong.
This is also the same switched speed I get between interfaces on the same bridge.
I have tried to use the switch chip and speeds increase to a more proper gigabit - but only when going through the same vlan!
Tomorrow I will be on site so I will be able to test the routing speed between two Routed interfaces that don’t go through bridges, but in the meantime I have a question
If I do everything without VLANs I just need to decide one master port per vlan and be done with it.
What if I have one VLAN that only exists on two trunks? Obviously I can’t choose a master port for them so I’d have to use a bridge. But in the Switch tab I get to create vlans and assign them to ports.
Can I create a VLAN on the switch chip, put two ports in it, and then give the router an IP on it and route it, while keeping switching speed at wire speed?
In Cisco language, an SVI. Without using Bridges.
This doesn’t address your question, but change the queue type to ethernet-default. You won’t get L3 at wirespeed but you should then get more than 30-40Mbps.
Very interesting. This already got me from 50 Mbps to 300.
Where’s the other 700 though? The switch is rated for gigabit routing as from its product page http://routerboard.com/CRS125-24G-1S-RM
Yet I am hitting 95% CPU while routing between two routed interfaces with no firewall rules applied.
Tested with iperf. If I do it completely switched I get 926 Mbps.
“AR9344 1G all port test” suggests, testing all ports at once [ie 24 routed interfaces] and reporting the aggregate throughput, although logically if you’re hitting 95% CPU testing through two ports @300Mbps, you’re not going to get another 700Mbps routing throughput with the remaining 5% CPU 
Note the steep drop off in performance as packet size decreases. Are you doing a TCP or UDP test with iperf?
Finally, note that as the CPU is connected to the switch module with a 1G link, and L3 is done in the CPU, then that’s your absolute limit on aggregate L3 throughput.
I’m doing tcp tests.
I am fine with this speed, I like this switch and for its price it’s a good thing already.
What I am complaining about is that writing “1gbps” on the performance tests does not seem fair if at the end all I get is 300 Mbps per port, after changing the queue types.
And I still wonder if there are better ways to do this
Bridging does not enable the wire speed switching and relies on the CPU.
Choose a master port and as many slave ports you want to create a group of switched ports. Traffic at L2 will switch at wirespeed between ports in that group