On Wed, Sep 01, 2021 at 08:08:48AM +1000, David Gwynne wrote: > I guess the overall question is if an IPv4 packet size is a hop by hop > thing, or whether the endpoints should have final say about what they're > prepared to deal with.
You are moving the layer 2 MTU size configuration to the clients. The router should handle this. Set the router interface MTU to the minimum your layer 2 network can handle. This would be 1500. If you don't like 1500 and know that paths to certain servers can handle more than 1500, you can set a more specific MTU at the routes of the router. The router will create fragments that fit into this layer 2 network. What you suggest, sounds like static path MTU. At every client you set the MTU to each server statically. Then you want the routers to keep the fragment size. That is a bit simmilar to IPv6 dynamic path MTU discovery, but without discovery. I don't think that your static path MTU configuration scales. That's not how IPv4 fragments work. I hope that I have understood you topology correctly. bluhm
