Hello, we ran out of time during this talk today and some comments were made at the microphone afterwards for which there was no time for follow-up. To address a couple of the comments:
Comment: Fragmentation is fragile on the open Internet Response: Yes, and that is documented in [RFC8900]. What this draft seeks to do is to make fragmentation robust within limited domains [RFC8799]. This draft therefore proposes to update [RFC8900]. Comment: Is the performance dependent on having the machines connected over a point-to-point link instead of a real network? Response: Increasing the transport protocol segment size increases performance *for that end system pair* regardless of how the segments are carried in the underlying Internetwork (i.e., either as whole IP packets or IP fragments) and regardless of the number of intermediate systems (routers). This is true even if the segment size exceeds the path MTU. Comment: Does a single transport protocol end system pair benefit at the expense of degrading performance for others? Response: All end systems should get better performance using larger transport protocol segment sizes, whether the segments are forwarded as whole IP packets or IP fragments. Comment: What does this do to routers? Response: The fragmentation and reassembly procedures are conducted by end systems; not routers. In the event that a router is required to perform network fragmentation, it sends an ICMP to the source (subject to rate limiting) requesting a smaller fragment size so that it will not have to fragment future packets. Routers otherwise forward IP fragments the same as for any IP packet. It would be interesting to receive feedback on these points. If there were other comments that I missed, or if I have somehow mis-characterized the ones captured above, please let me know. Thank you - Fred _______________________________________________ Int-area mailing list Int-area@ietf.org https://www.ietf.org/mailman/listinfo/int-area
