Even though DPDK sends packets down to the hardware as fast as possible, which is a great thing; it can also send down useful metadata. Some merchant NIC ASICs have hardware to support egress pacing, which their DPDK PMDs may or may not expose. Last year with our FPGA-based Arkville product, we developed a prototype "Paced Packet Player" which used TX/egress packet metadata to determine each individual packet's release to the MAC TX port, packet-by-packet. To ns precision. On four parallel 100 GbE ports. I blogged about it here.
http://atomicrules.blogspot.co.uk/2017/01/paced-packet-player.html The point is, that depending on your ability to add some dedicated logic just ahead of the TX MAC; DPDK's TX metadata can be used to provide egress packet-pacing to hardware. If or not your particular NIC can use this data is a function of your NIC hardware and the associated PMD. -Shep On Thu, Jun 8, 2017 at 1:51 PM, Vijay <[email protected]> wrote: > So, for example, at every 1 Sec interval if 1000 packets need to > transmitted, then HW tries to send them as fast as possible ? Is there any > document which talks about this ? > > On Thu, Jun 8, 2017 at 10:25 AM, Stephen Hemminger < > [email protected]> wrote: > > > On Thu, 8 Jun 2017 09:50:05 -0700 > > Vijay <[email protected]> wrote: > > > > > Hi, > > > > > > I am new to this mailing list and new to DPDK. I would like to know > about > > > packet spacing when a burst of packets are being transmitted. For > > example: > > > there is a timer running and at expiry of the timer there could > multiple > > > frames that need to transmitted. So, will these frames be spaced by > dpdk > > > before transmission ? Is it internally handled by dpdk ? Is there any > doc > > > that talks about this ? > > > > > > I am not sure if this question has already been asked. If yes, please > > point > > > me to the appropriate thread which talks about this. > > > > > > Thanks, > > > Vijay > > > > The packets in DPDK are not paced. The hardware sends them as fast as > > possible. > > >
