hello my friends in this forum…!
i built a bridge interface in my CRS326 and add all the 24 ports to it and check the HW for all port. so does that mean that the switch chip in my CRS326 is do it’s switching job and no load will be on cpu
or i have to do something else inorder to put this chip into services
In principle you’re all done.
Keep in mind even with offload enabled you will see some CPU usage mostly from Management tasks
but what a bout this switch section..? i mean if i can offload the traffic just by cheching the offload selection in the bridge ports, what is the benefit of the switch in this case..?
i mean the switch section in winbox .
You can configure some advanced stuff in switch submenu … one example are ACLs, another example is VLAN tag rewrites … I’m not sure these two are actually supported by switch chip in CRS326, they are on some of CRS3xx devices. But neither are supported by bridge logic.
Hi Techsystem
You need to enable l3hw on each port AND the switch itself. In fact disabling l3hw on the switch is the quick way to disable hardware offload in case of any issues (and there were many in the older firmware).
To verify if offload is working I would suggest running ‘/ip/route/print’ and confirm that you have the routes with ‘H’ flag. You can also do a traffic test between two nodes in different subnets using a tool like iperf3, and monitor the switch CPU utilization (‘/system/resource/monitor’) at that time.
l3hw is “routing offload” available on select devices (CRS3xx, CRS5xx, some CCR2xxx and perhaps some more) and is definitely not configurable via switch submenu - the functionality is far too complex to be handled by a few static settings. Indeed to enable l3hw functionality it has to be enabled in a few places, but device has to be configured as router and in certain fashion (not all different routing functions can be l3hw offloaded). But if things line up correctly, then such switch can route at wire speed (according to capacity limits, one such limit can be number of next hop hosts and directly connected LAN hosts usually count individually meaning large LAN can effectively disable l3hw offload).
Mind that l3hw support is depending on particular hardware and e.g. CRS326-24G-2S+ and CRS326-24S+2Q+ are pretty different in this regard.
The hardware offload of bridge functions and things in switch submenu are about “switching” … so frame switching between ports serving same (V)LAN.
Hello… so I’m thinking about “large LAN can effectively disable l3hw offload”…
There is a value that I can pay attention? actualy on crs326-24g-2s+ have 600 devices at /in/bridge/host print count-only, and my cpu is reaching 100%, so its already “large”?
The value which (in your use scenario) has to be observed is “IPv4 route prefixes” (and “IPv6 route prefixes”) from L3HW offload help page. This table contains IP prefix and MAC address of next hop. For explanation how the table entries are used, read note 1 under the table. And I guess that IPv6 table is problematic in particular as each host can have at least 2 IPv6 addresses (link-local and one or more global address). For crs326-24g-2s+ the table is pretty short (13k entries for IPv4, 3k3 for IPv6, somewhere in between for hybrid IPv4/IPv6 deployment) and I can imagine that 2k dual-stack hosts can already exhaust the table capacity.