> I'm suspecting the solution's in the network, not in AFS, however. We > just have to convince our network engineers (for the three networks the > packets cross) to believe that. :)
Looks to me that you jumped into the juboframe boat without checking the consequences at the border between the jumboframe - non-jumboframe world. There correct fragmentation must be take place or the sending part must ensure that depending on recipient, the right MTU is used. My guess is that the packet with DF is part of an MTU discovery scheme that is failing. (MTU discovery is generally failing in "the Internet" because people filter ICMP traffic). I'm quite sure that sending packets that are bigger than 1500 to any computer which in not on your LAN or where you have control over the whole path will only give you grief. So tune the OS/applications in that way. BTW, there are people who claim that jumboframes >4k are harmful as the checksum algorithm for ethernet is not very reliable for such big packets. You need to make sure that all interfaces on your LAN have the same MTU and that the routers at the borders do the right thing for packets that go between interfaces with different MTUs. ttcp (many implementations available) and tcpdump are your friends. Good luck, Harald. _______________________________________________ OpenAFS-info mailing list [email protected] https://lists.openafs.org/mailman/listinfo/openafs-info
