Behrooz, flow control allows a NIC to report the sender that it can’t keep up with the ingress speed. However this is just a hint for the sender (that it can ignore it) and being it at L2, it is just for a link. This means that if for reaching your peer you need to traverse several (L2) links, the flow control won’t work. In short you will have drops even with flow control on.
Luca > On 09 Feb 2015, at 16:45, Behrooz Shafiee <[email protected]> wrote: > > Hello all, > > When using pf_ring without PF_RING aware drivers (normal kernel drivers) > there is always chance of losing packet even though the flow control is > activated on the driver. Because NAPI might read faster than application and > pf_ring kernel module buffer overflows and drops the packets while kernel > backlog queue is not full (in case of a slow application). > > Now my question is that, if I use PF_RING aware drivers with flow-control > 'on' is there still chance of losing packet? I assume because there is no > queue involved in between (except driver rings) droping packet should not > happen, because if driver rings are getting full the drivers should send > pause frame to sender. Is that correct? > > Thanks, > -- > Behrooz > _______________________________________________ > Ntop-misc mailing list > [email protected] > http://listgateway.unipi.it/mailman/listinfo/ntop-misc
_______________________________________________ Ntop-misc mailing list [email protected] http://listgateway.unipi.it/mailman/listinfo/ntop-misc
