Hi Francis,
> Then there are two different points:
> - does the issue remain if edge routers are closed enough to hosts:
> * an interface only (for direct connections)
> * line speed limited at least by available stock hardware
> * low or at least reasonable number of flows?
In many cases yes.
> I believe the issue doesn't remain just because for many reasons
> stock hardware had, has and will have poor I/O performances
> - is it possible to put edge routers closed enough:
what I trying to say here is that people can build there network in
different way... as you know 99,99 % of the isp's still use IPv4 for
instance eventhough v6 is better and haver a nicer cleaner arhitechure..
> * technical constraints: I can't see one?
you will have to buy more memory which is very expensive since you must have
a recursive loop in the packet to find the offset where the transport header
sits - this is doable but much worse than ipv4 where you have the header
length and can dfind the offset immediatly for the transport header...
> * price constraints: with ASICs we are supposed to do it cheaply?
like I said it will be more expensier to build the hardware and in the end
it might have an impact on the rollout of ipv6...
> * policy constraints: I am sorry to have to say that but it
> is the job
> of a ISP to look at inside packets of its customers. It is why
> I said ESP will save us. Of course this should be the point where
> we have a very different opinion, in the past my current
> ISP decided
> the Web was mainly pornography and tried to limit the bandwidth
> for a particular TCP port...
And as I said I agree with your reasoning here from a pricipal clean IP
arhitechure but there are many service providers that wants to use this
feature.
> I have talked about firewalls because:
> - they shared the same issue but they have to do something a
> bit harder
> (BTW I think a solution which works for routers and not
> for firewalls
> will have a problem in the future: there are (too) many ISPs which
> put firewalls to protect (against) customers)
In other words we agree.
Like I said before layer 4 classification is not nescessarily tied to qos
classification it can be packet filtering as well and in both cases we have
the same problem.
> - they (will?) use ASICs when high performances are needed
not nescesseary asic it could be np's as well.
> - they share the management issue, security policies are even more
> critical than QoS policies (a mistake for instance has
> worse consequences).
I agree. But I think that the UML base QoS policy model for managing Qos
related policies is a very good start and hopefully we will have a security
base policy model as well...
Regards
thomas
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
