Frame Size Requirements
Page 18-12

MTU Handling

Routers address the problem of maximum frame size limitations with the notion found in
many protocols of a Maximum Transmission Unit (MTU) size. Protocols use this notion in two
possible ways.
• PDU Fragmentation/Reassembly
The router is configured with the MTU of each port. If a frame that is too large is required
to be sent on a port, the Protocol Data Unit (PDU) within the frame is fragmented into
many smaller PDUs, each of which is re-encapsulated and sent as a frame that fits within
the MTU.
• Connection-oriented end-to-end MTU negotiation
When an end-station enters into a protocol to communicate with another station the initial
PDU exchanges are guaranteed to fit all possible MTUs. In the handshaking between end-
stations to establish the connection a phase is entered where large frames are sent. If an
intervening link has an MTU too small for these frames it will be dropped and the hand-
shaking will time out. The end-stations send progressively smaller frames until the hand-
shaking succeeds and hence establish the MTU to be used between the two stations for
the remainder of their connection use.
IP supports the former mechanism and IPX the latter.

IP Fragmentation

The Omni Switch/Router Ethernet interfaces will use IP fragmentation if they are allowed to
(i.e., if the Don’t Fragment bit is not set.) Fragmentation by FDDI and Token Ring is not
supported though technically the Token Ring could send frames larger than those supported
by FDDI and LAN Emulation could generate frames larger than both.

ICMP Based MTU Discovery

IP uses the Don’t Fragment bit to support an MTU discovery protocol that superficially resem-
bles the negotiation of IPX. The difference is that when IP stations attempt to discover an
MTU size for their use, which doesn’t require fragmentation by intermediate routers, the
protocol expects a protocol response by the intermediate router, this is an ICMP reporting that
a frame was dropped because it couldn’t be fragmented.
The Omni Switch/Router transformation function of any to any switching does not support
this ICMP generation but just silently drops IP frames which can’t be fragmented. The IP
router in the Omni Switch/Router does honor this protocol and support ICMP. It is only the
any to any switching which doesn’t because it is not a router and may not even have an IP
address with which to respond.

IPX Packet Size Negotiation

For IPX the requirement of intervening devices is simply to drop frames that are too large to
be forwarded. This is what the Omni Switch/Router does.

Other Protocols

Dropping oversize frames is the approach for all protocols other than IP. If the protocol in
question is modeled like IPX this will be the correct thing to do and will not cause problems.
If the protocol is modeled like IP and expects fragmentation to occur or requires explicit
response from the Omni Switch/Router then the protocol will not succeed in any to any
switching.