I have installed 7.1.3 to CRS318-16P-2S+. I tried to setup the L3HW interVLAN routing based on the website documentation. I think it doesn’t work because I always got high CPU percent.
Is it possible on this device?
(I tried to enable l3hw on the ether1 also, but doesn’t work…)
And for inter-VLAN routing you also need
/ip firewall filter
add action=fasttrack-connection chain=forward connection-state=established,related hw-offload=yes
NOT all CRS 3xx switches support Offloading Fasttrack Connections
devices based on Marvell 98DX224S, 98DX226S, or 98DX3236
only support support L3 HW Offloading, do not support Fasttrack or NAT connection offloading.
CRS318-16P-2S+ is based on 98DX226S
CRS318-1Fi-15Fr-2S is based on 98DX224S
Offloaded Fasttrack connections (applies only to certain switch models)
Offloaded NAT for Fasttrack connections (applies only to certain switch models)
hi, CRS328-24P-4S+RM, after reboot device no load default route to hwl3. Route is in routing table with mark DAoH but not function. After off/on /interface/ethernet/switch set 0 l3-hw-offloading default route is function. Same with CRS326. With the CRS326, this happens after a few hours of operation.
Thanks hapi, the same problem is reproduced in our labs and we look forward to fixing it in future RouterOS releases. No release date available at the moment.
We have found a few semi-severe issues on 98DX2xxx/98DX3xxx switch chips (CRS305, CRS318, CRS328, etc.) when L3HW is used together with dynamic routing protocols (OSPF, BGP). Most of the issues have already been fixed in v7.2rc5. The remaining ones will be fixed in the next version. Unfortunately, fixes will not be backported to the v7.1 branch, so you must switch to 7.2 to use OSPF/BGP with L3HW.
Routes received via BGP were not properly installed to switch chip on CRS328-4C-20S-4S+ with 7.2rc5.
Symptoms:
establish a BGP session, connectivity is normal
enable L3-hw-offloading, connectivity still normal
restart BGP session, connectivity is lost until L3-hw-offloading is disabled.
Is there a point in upgrading to rc6, i.e. was something like this addressed?
Something surely changed for the better with rc6, however I’m uncertain what fixed things. There was never any lost connectivity post-upgrade, however packets were routed by CPU. This was fixed by adjusting vlan interface’s MTU (port had 9000, yet I didn’t notice that vlan had 1500). Tests seen 3 gbit/s with cpu load at 1%, so as long as things are stable, I’m happy. Good job guys.
Im trying to enable L3HW routing on some 317s. They have 900 (all incl 0.0.0.0/0 from OSPF) routes and right now I tried only with switches having one uplink.
When I use
/interface/ethernet/switch set 0 l3-hw-offloading=yes
/interface/ethernet/switch/port set [find] l3-hw-offloading=yes
The traffic seems to flow fine for the users but I loose local loopback reachability outside for things like routeros upgrades.
Same on 7.2rc7 so it might be one of still not solved usecases?