Do not fragment

Do your own homework. Ask your teacher.

- First : fragmentation : Not all networking technology have the same frame size. Case in point Ethernet and Token Ring. The MSS (Maximum Segment Size) of an Ethernet network is 1500 Bytes. The MSS of a Token Ring network is 4500. The MSS of some other networking technology is only 476 bytes.

So a router transferring a Token Ring frame to an Ethernet network need to cut that packet in roughly 3 parts. This is why you need to do fragmentation. Now, since transferring a TCP packet onto that frame require some headers the number are actually smaller.

Routers will fragment packets too big for the next step on the network path. Only the destination host will (except for IDS/IPSes) have to reassemble all the part to make it whole and ordered again. In the case where the network is not very efficient and loose packets badly, you will miss parts and your communication will have to manage timeouts and ask for retries.

All that take time and computer resources. In some cases, it's better if there is never any fragmentation.

As per in our example, if the 4500 bytes Token ring packet goes to a router with the DF flag set, then the router will send back to the originating host an ICMP type 3 with code 4 (Destination Unreachable ) and (Frag needed and DF set.)

The maximum size of a TCP segment or UDP datagram is 65565 bytes minus the header size of the IP packet header and the header of the TCP/UDP part.

Since TCP is more complex than UDP its header take more space.

All of that is is clearly written in the specifications of the protocols in documents called RFCs. All RFCs can be found at :

Well I hope it help.

By the way if I may correct the vocabulary for the name of a header and a data payload at OSI layers

At layer 1 (physical) it's called a bit stream... or serial comm. At layer 2 (data) it's called a "frame" (Ethernet) At layer 3 (network) it's called a "packet" (IP) At layer 4 (transport) it's a "segment"(TCP) or a "datagram"(UDP) Want me to go further ?

And at layer 8 (human!) it's called correct vocabulary...

What exactly do you mean by "process it" and by receiver do you mean the actual end destination, or an intermediate router?

Details about IP and ICMP processing can be found in their respective RFC's which are archived at

Approximately the size of New Jersey, or the length of a piece of string. :) The UDP RFC(s) are also at

stricly speaking (at least as strictly as I can recall)

Application "packets" are called messages TCP "packets" are called segments UDP "packets" are called datagrams IP "packets" are called datagrams Ethernet "packets" are called frames

Sooo we can have for example an application-level message carried in a UDP datagram that is carried in some number of IP datagram fragments, each of which are carried in separate Ethernet frames. If that is then set over an ATM network those frames are further bend, folded, spindled and mutilated into cells.

rick jones

I've always understood that IP packets were called datagrams.

We were all new at this once. Just that as we get old at it it becomes harder to remember what it was like to be new. To layer is human, to tolerate divine :) :) :)

Oh, and layer 8 is actually the Financial layer:

rick jones