From: Tony Li [mailto:[EMAIL PROTECTED] 
>In some ways (and I admit that this is an imperfect 
>analogy), map-&-encap asks our hosts to trust their 
>transport to a technology that they can no longer 
>interact with.  How do we manage the underlying network?  
>What happens when a host needs some exceptional 
>service from the network?  This seems problematic 
>and complicated, when the root architecture of the 
>network must be truly simple and elegant.

How can hosts make requests of the network today? The only direct
mechanism that I am aware of is QoS. The only other mechanisms that come
to my mind are out-of-band systems such as application-layer admission
control systems -- the latter often rely on ECN reporting. Cumulatively
these systems rely on the TOS-byte of the IPv4 header and the IPv6
equivalent. What am I missing? How does anything change with
map-and-encaps? To my mind, there is no change. What are you seeing that
I am apparently missing?

For example, if your reply to my first query (above) mentions MPLS then
we will disagree -- I know of no host based systems that use MPLS. I am
only aware of network services using it (e.g., L3VPN). It wouldn't
surprise me if people try to create host services that use MPLS. The
problem with doing that is that it doesn't scale. More to the point,
such systems create configuration headaches for the large end user,
which is A Bad Thing. 

I've also seen deployments try to use the IP header's source route
option (e.g., to differentiate between gateways). That's another
technique that doesn't scale. 

Yet even so, both approaches (MPLS, Source Route) don't change with
map-and-encaps because they can only know about their own recursive
instance (i.e., our hosts are not be aware that our ISPs (note the
plural) internally convey our traffic across their infrastructures using
MPLS).
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to