I voted for (b), with our past experience.reason.
It may be useful to comment from out experiences, so I'll try to
explain.
We had some offsite meeting network setup in Japan, just like IETF
terminal cluster, but with more aggressive experimental one. We used
diffserv with prioritized queueing of some sort ([EMAIL PROTECTED]
should be able to refresh my memory about more detail), as well as
integrated MPLS + diffserv network (diffserv classification, you can
purchase bandwidth guarantee by virtual currency, traffic goes through
dedicated MPLS path). Both of the time, major portion of the traffic
was HTTP (TCP port 80), or SSH (TCP port 22). the fact made the
classification guys (including kjc) irritated, as almost all the
traffic look the same and there's no fun in doing port-based
classification. Also, once someone uses ESP, it becomes impossible to
classify based on port number.
So, kjc wanted to at least identify each of the flow, at least for
statistics purposes, as well as traffic classification/policying
if possible. Even if end nodes use ESP, he would liked to identify a
flow from other flows. So, he wanted us to attach unique pseudorandom
flow label per flow. This leads us to (b), and that's one of the
reason why draft-itojun-ipv6-flowlabel-api-02.txt is based on (b).
It is true that the originating node can put random value into flow
label and confuse classifier, but there's no point in doing that.
Also, it is true that it may help if the originating node can fill in
port number information into the flow label, however, the originating
nodes are not that cooperative. End-to-end pseudorandom value
looks like the most useful, and most easier-to-manage for originating
nodes (without requiring too much information leak).
itojun
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
