[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: geo short vs long term? [Re: Geo pros and cons]
On donderdag, apr 17, 2003, at 10:29 Europe/Amsterdam, Kurt Erik
I think the first step still is to define what the options are. Then
we can start worry about consensus.
In any multiaddressing scheme, there must be a point where a single
"handle", be it an address, an FQDN, a control block or what have you,
is tied to two or more paths through the network. There are many ways
to do this:
- let the applications handle it (SMTP, DNS)
- let the network layer handle it (MIP)
- let the transport layer handle it (SCTP)
- let a middlebox or border router handle it (MHAP)
GSE isn't easily classified here as transport, network and border
routers all need to play their part.
I think we can agree that having the applications do it isn't good
enough for many applications as sessions don't survive a rehoming event.
Network layer has the advantage it is presumably easy to implement.
Transport layer has the advantage that it already has the end-to-end
state which makes failover easier and doesn't pollute the architecture
by introducing state elsewhere.
Middlebox/border router has the advantage the number of boxes that need
to be upgraded is very small.
These are our basic choices. The problem is that each of the advantages
and disadvantages will be appreciated differently by different users.
Enterprise users want central policy control, home users and mobile
users want flexibility no no reliance on provider cooperation. Some
people need multihoming in IPv6 yesterday, others haven't even heard of
IPv6 in the first place.
If possible, I think the best choice is not to choose at all, but make
the mechanism generic enough that it can work anywhere.