brescia@PARK-STREET.BBN.COM (Mike Brescia) (09/22/88)
> Is there an agreed upon max datagram size for UDP packets ??
There is an assumption at the IP level that says an IP implementation does not
have to be prepared to reassemble and buffer individual packets of up to 576
octets. This is listed in RFC791 (IP protocol spec) in the description of the
IP header Total Length field. "It is recommended that hosts only send
datagrams larger than 576 octets if they have assurance that the destination
is prepared to accept the larger datagrams." This assurance is only provided
by specifications in higher layer protocols. Some IP implementations are
capable of reassembling up to the full 64K datagram size without prior
agreement.
For TCP, there is an exchange at the time of opening the connection where the
2 parties can agree on datagram sizes larger (or smaller) than 576, using the
TCP MSS option. Two hosts on an ethernet may well negotiate for 1024 data
octets.
In answer to your question about UDP, I would have to say that, yes, there is
an agreement on datagram size, but it is specific to the higher layer
protocol. TFTP uses 576. Sun NFS uses 8K. I think you would have to
implement the UDP part of a system so that it would be able to handle whatever
a higher layer would want.