> > I think that "messages should be regular in format and easy to parse" is a
> > reasonable requirement, though. This might reflect into fixed-length
> > fields, or labelled fields, in the protocol specification.
This also will reflect in a low probability to commit coding error as
buffer over flow or DoS problem even if maybe a fixed length
protocol is more adapt for UDP and a labbelled fileds is more adapt for TCP.
If the average message length size is near to a big percentage of messages len.
the fixed lenght field protocol may be more acceptable but add an intrinsic
limitation. Also it's possible to reduce the overhead using a common way to
perform both strong auth and cipher. For example it's possible to add two
160 bit fields, first with a truly random number, second with an hash of
(secret+msg body), altered random number+secret may be used to build
a stream cipher using only hashing and xor. But it's only an example.
antirez
