Alex Bligh <amb@gxn.net> writes:

LINX also runs mixed media (sadly) - FDDI and more sane things (GigE and Fast Ethernet). The switches here fragment when switching toward Fast Ethernet. They would certainly move >1500 byte pings between FDDI ports, but I have no FDDI ports now to test whether this works with DF set (i.e. it doesn't fragment FDDI to FDDI), nor whether ICMP is *ever* generated.

Maybe I'm being ultra-thick here, but what business do layer-2 devices have looking at the contents of layer-3 packet headers and acting on what they find there? I thought the whole point of building IXPs at layer-2 was to leave layer-3 policies to the participants - if not then one might as well start building IXPs with routers surely?

However, I believe best practice at LINX is, and perhaps should be elsewhere, to set MTU on your FDDI interfaces to that supported by ethernet. Given most exchanges started with ethernet and worked upwards, this should have been best practice everywhere from teh start.

This seems to me to be exactly the right thing to do in the context of exchange point fabrics composed of mixed MTU media, irrespective of the presence or absence of such schizoid layer-2 devices as those that have been discussed here. M.