Regarding the switching capabilities of the 10-port RB3011:
With the addition of hardware offloading since RouterOS version 6.4.1, if ports from both switch groups are connected (bridged?) into a single switch, will they all communicate at “wire-speed”?
For example, if I use two ports for other purposes, then have all eight remaining ports configured as a single switch, will all eight be working at wire-speed (no CPU involvement)? Or, will it still be slower than the 5 ports in the basic switch group?
How does the hEX S (RB760iGS ) compare in this regard?
Ports communicate at wirespeed between ports of the same switch (1-5 and 6-10) but not between ports of the two switches.
So with some clever arrangement of systems on the ports you may be able to get it to work satisfactorily (i.e. do not put all your servers on one switch and all your clients on the other).
When you have wirespeed switching requirements for a lot of ports it may be better to use an external switch for that.
If you need 5 or less ports - just use ports from one group for the LAN. (WAN, if it’s just routed, without bridged VLAN, can be in the different group then)
If you need 9 or 10 ports - then, unfortunately, you need an extra switch to work with wire-speed without using the CPU.
For 6-8 ports… You can always:
Configure 2 separate bridges (ports 1-5 and 6-10), each with hw-offload
Configure everything (VLANs, IP addresses, etc.) on the first bridge
Connect ports 5 and 6 with a short cable
Now you have ports 1,2,3,4,7,8,9,10 working with wire-speed (also with VLAN tagging/untagging)
fast-path and fast forwarding can get a lot of performance, if you can live with a restriction of 1gbps between the 2 switches do it by bridging maybe you can get more than that, and getting 10 ports
Of course it doesn’t get actual wirespeed. When you would have a single 10-port switch you could have a 1 Gbps
transfer between port 1 and 10, and another one between 2 and 9, and another one between 3 and 8, all at the
same time. But with this solution you can have only 1 Gbps between the lower and the higher port group.
Indeed it could well be that the bridged solution outperforms this, because the link between the switch and CPU
is 2.5 Mbps and so is faster than each port.
The advantage of the wire method is that it does not load the CPU. Not that it is wirespeed.
Between physical ports you can not exceed 1Gbps/port and direction, because the ports are 1Gbps .
The CPU ports have a total forwarding capability of 2 Gbps in each direction…
So, cross the switch groups you could get somewhere between 1Gbps with no other load and 400Mbps on full switch load (assuming fair queuing, no SFP plugged and no other CPU load) per direction, but in real life you will probably not hit this limit, since you probably will not use it at the core of a data center.
Hi, I know this post is quite old, but I have a RB 1100 AHx2, with a bridge in ports 1 to 11, and in port 13 the WAN
If I build a switch, how do you explain here will I win in performance? Is this design better than a traditional bridge?
Smilar story: RB1100 AHx2 has two switch chips, one has ether1-ether5 attached and the other ether6-ether10. Interfaces ether11-ether13 are separately treated and directly attched to CPU.