[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [RRG] On "jack-down" models
Dino,
|Tony, that is what we recommend in LISP. So we don't have to re-
|address the core. We force the re-addressing of sites that don't
|currently have PI blocks. We also are loose about stating that sites
|can use PA blocks for EIDs. But with the later, the namespace is not
|mutually exclusive. That is okay, if the two never cross.
Thanks for the correction.
Your welcome.
|> Handley
|> posits the
|> use of multiple parallel connections between hosts, striping data
|> across
|> these connections to instantiate a single, address-agile transport
|> layer.
|> Implicit in this structure is a mechanism for the host to recognize
|> that
|> these individual connections are part of a greater aggregated
|> connection.
|
|You get that as well when middle-boxes load-split ingress and egress
|traffic. The transport connection in the host just sees a
|32-bit value
|as a connection ID.
Ok, but you don't get to take advantage of the parallel paths
between the
end host and the middle box(es).
Well that is an IGP policy that is already implemented in most sites
today. You can ECMP from each IGP router on 0.0.0.0/0 to each ITR. So
that problem is already solved.
And LISP doesn't change that.
Think of all of the time and effort that folks have put into ECMP
routing.
Doesn't it make some sense that they will want to eventually
propagate this
all the way back to the host?
It is propagated to the first-hop router. In fact, we are even adding
techniques to get shortest paths from a source to the first-hop router
when there is a multi-hop switch network between the two.
|I don't see this mechanistically any different than Shim6 or Six/One.
|I look forward to more details of Mark's proposal so we can see the
|differences.
If I understand it, there's a drastic difference. First, all of the
multihoming issues are dealt with by layer 4 (and possibly above).
Individual transport connections would just use vanilla v4
addressing as
it's used today, complete with using the address as identification
for the
component connection. These would then get tied together by a stat.
muxing
layer.
That is a shim6 approach IMO.
That would leave routing completely and wholly unchanged, and allow
upgraded
hosts to migrate to pure PA addressing without issues. Upgraded
hosts could
still interoperate with legacy hosts using legacy transport
mechanisms.
In LISP routing is left unchanged in the site and in the core. The
mapping function is what is added and only in the CE boxes.
Routing is really not changing. If it did we couldn't do the LISP
pilot over the Internet right now. And many of us are doing this pilot
in a corporate environment where the IT department has no idea what we
are doing, hence no site changes.
Dino
--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg