Hi,

What's the problem with mirroring? It increases CPU usage since packets enter the CPU. However, unlike the software routing, the CPU does not throttle the performance in the case of mirroring. For instance, if the CPU can handle 10G while the actual traffic rate is 50G, all 50G will get hardware-routed while only 10G gets mirrored to the CPU (the rest is tail-dropped from CPU queues). But if you need to monitor 100% traffic, then I see no other way but disabling L3HW and throttling bandwidth down to the CPU capabilities.

There is no solution for this. L3HW offload implementation doesn't provide any detailed statistics, so if you need it, then ... well, you unfortunately can't use L3HW offload.

Hi raimondsp

Hardware level traffic sampling on the Marvell ASIC's to netflow/sflow would be the optimal solution, to quote you "Hardware traffic sampling and QoS are the next major features that we consider implementing after the finalization of IPv6 L3HW. "

Is this still on the high priority list ?

It will allow exporting of flow information to DDoS systems well as well as flow accounting systems like nfacct that are widely used.