Hi Mikael, > -----Original Message----- > From: Mikael Abrahamsson [mailto:swm...@swm.pp.se] > Sent: Friday, July 27, 2018 1:21 AM > To: Tom Herbert <t...@herbertland.com> > Cc: Templin (US), Fred L <fred.l.temp...@boeing.com>; internet-a...@ietf.org > <int-area@ietf.org>; intarea-cha...@ietf.org > Subject: Re: [Int-area] WG Adoption Call: IP Fragmentation Considered Fragile > > On Thu, 26 Jul 2018, Tom Herbert wrote: > > > I don't understand what would be platform specific. Can you elaborate? > > On one platform iperf3 sends MTU minus header sized UDP packet, without IP > fragmentation (so ~1400 bytes UDP payload). On the other platform iperf3 > sends 8k UDP packets, fragmented by the IP stack if MTU is less than 8k. > > I have no idea if this is an iperf3 version difference or platform > specific, but this is what I see in my tcpdump:s.
Out of curiosity, what is the performance being reported by iperf3 on iOS? On linux, the performance for sending 1400 byte UDP datagrams is less by an order of magnitude than sending 8K UDP datagrams even though the 8K's incur fragmentation. Can you try boosting the UDP datagram size to 8K on iOS (i.e., and let it fragment) and see what performance is reported? Also, how does performance on iOS compare to performance on linux for similar network configurations? >From my testing, the reason for the big performance disparity on linux is the size of the buffer being offered to the "sendto()" system call. Offering small UDP datagrams means that only a small data copy occurs per system call. Offering large UDP datagrams (and therefore larger sendto() buffers) yields greater performance even though IP fragmentation is exercised. Thanks - Fred > -- > Mikael Abrahamsson email: swm...@swm.pp.se _______________________________________________ Int-area mailing list Int-area@ietf.org https://www.ietf.org/mailman/listinfo/int-area