HW Offloading

Hi, I have an issue on a CRS309 where when I enable HW Offloading on the WAN connection (sfpplus-1 in the attached screenshot), traffic will no longer route. The route shows the ‘H’ and HW Offloading is successfully enabled and shown as working on all routes, but the connection effectively stops working. I have attached some screenshots of the config in winbox (without HW Offloading enabled). I do have NAT enabled on that interface. The HW Offloading does work on the local lan with no problem.

I’m thinking there is some limitation that I’m not aware of, or I’m just missing something. Would appreicate some assistance.

Cheers
Screenshot 2024-03-27 000947.png
Screenshot 2024-03-27 000451.png

L3HW offloading only works between if all routes reside on same bridge. It seems your WAN is on off-bridge interface sfp-sfpplus1 .

This requirement applies only to VLAN. If the WAN port does not need VLAN tagging, it can stay standalone - L3HW will still work.

I guess the problem is with NAT. Please read this:
https://help.mikrotik.com/docs/display/ROS/L3+Hardware+Offloading#L3HardwareOffloading-Inter-VLANRoutingwithUpstreamPortBehindFirewall/NAT

Short answer (from what Raimond pointed to):

For NAT to work, the traffic has to hit the CPU initially, and then it pushes those NAT sessions to the ASIC. You therefore can’t enable L3HW Offload on the WAN ports or else it’ll never hit the firewall. You can only enable it on the LAN ports.

Just to be clear is HW offloading possible on some routers regarding its chip, completetely different from L3HW offloading discussed for switches?
I am trying to make sense of test results.
For routers, I look at the table with or without filter rules and assume the speeds reflect the best possible results of the router, including HW offload enabled if applicable ( like RB50009 )

In the case of switches I do the same.
However folks here are indicating that L3HW offloading on switches renders ethernet test results as being completely bogus.

RB5009 doesn’t support L3HW offload.
On routers that do (those have capable switch chips built in), the L3GW offload concept is the same as on switches.
The difference is in the effectiveness of handling traffic which for some reason (e.g. route prefixes already offloaded use up all the ASIC route prefixes table capacity) doesn’t qualify for offloading and has to be handled by CPU (routers obviously do a way better job here than switches).

Some devices have published also test results for L3HW (e.g. CCR2216-1G-12XS-2XQ). Numbers are impressive (IMO not bogus really :wink:).

also will be nice to know which version of RouterOS is Running in your CRS309

Very nice MKX, you missed the boat and the barn, as we are talking about switches, and you bring up a ROUTER, not a switch.
None of the CRS3XX series of switches then has L3HW offloading if I had to base it on ethernet test results ( very slow ).
So once again I am searching for some truth and facts.

Didn’t somebody mention routers a few posts higher?

Generally I don’t really trust test results from MT. So in this case I’d go with documentation, like official L3HW offload manual with its L3HW Device Support section.

Yes I know HW offloading is available on some routers and now edumecated that L3HW offloading is also available on at least one router.
But it just proves my point, you were unable to answer the thrust off this and may other threads where people are using switches to do routing and wondering why L3HW unloading is not providing fantastical ethernet results. No one cares about L3HW offloading on routers LOL, unless you are a 2216 owner living the dream…

I will see if there is more uptodate L3HW offload video available… I just want to make sure advice is accurate and not a guess or wrong…