[Seamoby] RE: what is edge signalling? (was: RE: [NSIS] Re: Comments on dra ft-westphal-nsis-qos-mobileip-00.txt)

["Gary Kenward" <[e-mail]>]

Robert:
 
  I will have to read Cedric's draft. I wasn't aware that it discussed CT.
With deference to Cedric, and what I am sure is a thoughtful
analysis, it seems to me that understanding the relationship between CT and
NSIS would require, at the very least, a CT solution, or
something close to a CT solution. Clearly there is a difference of opinion
as to what CT should be (the requirements draft is apparently
going to go through another revision, so who knows?).
 
  In my view, if CT is in place, then NSIS should not be invoked as a result
of a handover. A pretty simple relationship. The only exception
that I can imagine is if there is some impact on a service that is not
provided at an ANR (i.e. an AR or an internal router), then NSIS may
be required to maintain that service. However, I could not name this service
at this point.

-----Original Message-----
From: Hancock, Robert [mailto:[e-mail]]
Sent: July 5, 2002 10:59
To: Kenward, Gary [WDLN2:AN10:EXCH]; '[e-mail]';
[e-mail]; [e-mail];
[e-mail]
Cc: [e-mail]; [e-mail]; '[e-mail]'
Subject: RE: what is edge signalling? (was: RE: [NSIS] Re: Comments on dra
ft-westphal-nsis-qos-mobileip-00.txt)


hi gary,
 
my comments are purely on draft-westphal-nsis-qos-mobileip and associated
email discussions (since that's been the only input I have seen recently
which explicitly discusses the relationship between CT and NSIS, and which I
had worries about). 
in particular, any comments I make about MN-AR signalling are really to
emphasise the difference between the cases of
a) not having CT signalling that carries QoS information MN-AR (if there is
or isn't other CT signalling I don't really care), and
b) having CT signalling that does carry QoS information MN-AR, which seems
to be the assumption of the mentioned draft.
 
our recent framework activities assumed (a). if you're saying that the
seamoby w.g. requirements also impose (a), that's fine by me. i also don't
want to re-open old discussions (certainly not on the nsis list, anyway).
 
i suspect we're basically in agreement about NSIS/CT relationships: after a
handover, if CT happened, it should be possible to have less NSIS signalling
on the on-path segment which goes over the air.
 
cheers,
 
robert h.
 
PS referring to a previous email, i think the view that NSIS is "likely to
be of limited use in mobility situations" is rather pessimistic. i would
rather say that if you want the resources signalled for by NSIS to be
available seamlessly over handovers, you'll need something else as well,
which could be CT. the basic point seems to be that NSIS shouldn't try to
solve the cross-path context/state transfer problem - any more than CT
should try to solve the along-path context/state establishment problem. 
[gwk] I agree. I was not trying to be as much pessimistic, as I was pushing
back against this concept that NSIS + LMM => Seamless Mobility. To be more
precise,
if it takes longer then a few 10s of ms to set up the new, conditioned path,
then the handover will likely not be seamless in 3G. Signalling,
particularly over the 
air (again, 3G), takes relatively significant amount of time to complete.
10s of ms is a rough approximation, but if you are running voice frames,
more then, say 
20ms of additional handover delay will start  impinging on the service
requirements for the voice - i.e. either frames have to be dropped or
delayed beyond 
the performance constraint. It is difficult to pin down these numbers,
particularly in an IETF email list, but any delay budget analysis that I
have seen usually 
yields numbers in the stated range (keep in mind that delay/drop allowances
have to be budgeted for other components of the access network and for the
link layer handover).
 
NSIS + LMM (or some variant of MIP) might be one approach to mobility. Fine,
but given what I understand about what NSIS is, and is not, it can never be
any where near as seamless as a solution that incorporates CT. Is CT simply
an "optimization" - well, that depends on how much service disruption the
user
can tolerate. Usually, there is a threshold where the service is simply
considered unusable.
 
Cheers,
Gary