[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
traffic type header
The suggestion has been made that we should have an optional traffic type header.
That is, we need something that will disclose protocol and port numbers. The
purpose would be (a) to permit traffic measurements, and (b) to permit routers
to optimize their handling of certain kinds of traffic. (And yes, it's optional,
in case traffic analysis is your foe.)
Neither IPv6 flows nor IPv4 traffic types are a substitute. Apart from the fact that
the former isn't here yet and the latter isn't used, they just don't disclose
enough information. But if you're engineering a network, you need to know these
things. To give just one example, the amount of http traffic to particular places
can be used to justify or refute the need for an organizational Web proxy.
As a strawman proposal, let me propose the following header format:
u_char next_proto;
u_char this_proto;
short pad;
u_short src_port;
u_short dst_port;
Arguably, pad should be length, in which case we could have two optional
fields:
struct in_addr src_host, dst_host;
to account for tunnel mode. (Tunnel mode is useful even if you're not trying
to hide the host names.)
An alternate mechanism would be to define new transforms; if nothing else,
that would avoid adding yet another header to the protocol type namespace.
It would also be simpler to implement; the obvious ways to do this in a
BSD-derived kernel or a streams-based kernel just don't work very well.
If there's enough interest (and, of course, not too much opposition), I'll
volunteer to write a draft RFC.