Hi everyone! Hope someone can help me understand an optimal config for the RB4011 - specifically with a 10GB SFP trunk connected.
Based on the block diagram below;
In the default out of the box config, Eth1 is allocated for the external routed connection and Eth2-10 and the SFP are included in a single bridge.
#1: My interpretation is that for L2 connections in that default bridge;
a) Any traffic passing between Eth2/3/4/5 (ie. same switch chip) wont touch the CPU and will leverage the hardware offload for those ports
b) Same for traffic between Eth6/7/8/9/10 hardware offload bridged but only between those ports
c) Although they are technically in the same bridge, traffic passing between these two hardware groups (e.g. from Eth3 to Eth8) will have to traverse the 2.5GB CPU lane and incur some CPU overhead
d) Same for the traffic coming from the SFP port - even though its a 10GB connection, data passing between say a downstream CSS326 switch anything in the other ports in the default bridge will have to hit the CPU
Oddly when I look at the bridge it says that all ports, incl. the SFP are hardware offload enabled (HW=Yes) so from the block diagram I’m not sure how that is? Unless I’ve misunderstood how the hardware offload works? Or is there a hardware offload inherent in the SFP that isn’t described (or is it just not relevant for those high speed connections).
#2: For L3 connections (in/out the routed Eth1) connection they’re pretty much all the same - they’ll always hit the CPU lane, and assuming the usual caveats - will get fasttracked. Regardless of which switch chip or port they are on.
So 2x Questions for clarification are;
Q1: if 99% of my bridged/LAN clients are on a downstream switch on the SFP port, there’s no real need to incur the overhead of bridging in those additional ports - and I might as well just drop the bridge and use the SFP as a standalone port, keep all the L2 traffic downstream and fasttrack where possible through the CPU to Eth1.
Q2: Alternatively If I keep the bridge as it is, and have some devices that predominantly route externally (as opposed to being LAN-heavy but still require some local network broadcast access), theres no real benefit to elevating them to Eth2-5 because even though they’re on the same chip, they still have to pass through the CPU lane , even if its only to get quickly fastracked back out. This will likely reduce traffic on the downstream switch, but will incur a higher CPU overhead due to the bridge straddling the CPU.
Thanks in advance! Just trying to make sense of whether the default bridge config is optimal or just easy for ‘plug n play’ reasons.