RE: framework models, related architectures and protocol controversy
Moisand, Jerome <jmoisand <at> juniper.net>
2002-12-04 23:34:16 GMT
Melinda,
yes, I don't disagree with you, scenarios where a middlebox of some sort
would have to be dynamically provisioned for a given TCP/RTP session are
making VERY SPECIFIC assumptions about the IP topology. It looks to me that
the only realistic scenarios are at a domain boundary, typically the point
where a service provider provides IP services (e.g. a WAN uplink). Then,
either on the service user side (e.g. a NAT/FW CPE device) or on the service
provider side (e.g. a BRAS device), the topology is constrained enough that
such scenario makes sense.
This being said, such specific scenarios are so important to both service
users and service providers that I believe it's really the pain working on
such middlebox/dynamic-session-management stories. And as I pointed out, the
whole picture involves filtering, NATting, firewalling and Qos
differentiation.
As to scaling, I hear the concern, but I'm not sure to entirely agree with
you. There are solutions nowadays which can scale pretty well by the
appropriate combinations of hardware & software.
Finally, on the "protocol controversy" wording, I apologize if my wording
was a bit too aggressive. I have to say I'm a bit tired of the endless
discussion between SNMP, Diameter, COPS across multiple IETF work groups. I
tend to believe that all these protocols have a role to play, and we'd
better be flexible, and define MIBs, PIBs and Diameter attributes for
multiple functional areas (ok, when this makes "reasonable" sense, of
course, which is hard to define!), instead of spending too much time arguing
in an often religious way. I hope that I'll not get a flame on this
(Continue reading)
RSS Feed