NSIS IETF#68 Meeting Minutes
============================

Meeting Minute Taker: Andrew McDonald


agenda bashing (Martin)
 - tunnel draft moved to end - not got anyone to present
 - behind on milestones
 - qspec template wglc completed - revision needed
 - qos & natfw still in progress
 - rmd in wglc from 15 march to 5 april
 - mobility draft - takako taken over as editor
 - gist protocol implementation document
 - extensibility model - expired individual draft but referred to by gist
   needs more contributors
 - 3gpp2 natfw draft will be progressed as AD sponsored document
 - interop in Karlsruhe?
   christian dickman: put together an initial doc on test cases
   some discussion off-line
   will post to list once half-ready

gist status:
iesg current state: 1 yes, 5 no objection, 4 discuss, 5 abstain
need 2/3rds yes/no-objection

returning to working group for wglc, ietf lc, ...

henning comments of two forms:
comments on draft
comments on unfixable problems (nat, etc)

rsvp would fail more badly than gist if given such a review

do we need a document that discusses some of these issue to aid
the last call (nats, flow splitting, etc)

additional problem is that quite a few iesg members have changed

don't want to make gist draft longer - already long enough

lars eggert:
the iesg members with discusses remain
quite a few of the abstainers replaced
so that may help
not common to send back to working group
quite a strong message from iesg

henning:
discusses are from people who are willing to engage in technical discussion
bigger problem are the abstains
need to particularly look at early pages of draft
better than early versions

henning:
would it help to put together a "tutorial" to help these people
get the big picture

lars:
there are multiple implementations, so interop is not an issue
the problem is that different people worry about "how does it
interact with /this/" where different people have a different /this/
depending on their area of interest

rob:
lots of these things are ill-defined

-- 

gist v12 - robert hancock

protocol changes, technical clarifications, editorial changes

fair amount of extra text added to early parts of draft to clarify
what it is about

legacy nat handling (pc5):
new section 7.2.1
base specification now says: here's how you detect a nat and then give
up open to extension specifications to define better handling
old draft was silent on what happened

q-mode encapsualtion (pc6):
avoid intercepting packets to gist port which aren't gist
added magic number
still an open issue related to this
also tc13 discusses what has to be done with ip options

state machine updates (pc7):
mostly from comments by adrian farrell
better handling of "no routing state" error messages
better/simpler lost confirm handling

routing state authorisation and identity checking (pc8/9):
wire protocol not changed, but what needs to be done in implementation
is clarified - rules on identity checking now given

rao handling (pc10):
comments from adrian farrell - routing area review
clarifications related rao allocation
absolute prohibition on q-mode fragmentation
new informative appendix c on what might go wrong with rao

side point related to extensibility draft:
when to request new rao values

transport properties:
tc1/tc2
change 'MAY' use C-mode to 'SHOULD' unless capacity is engineered

path splitting and route flapping (tc19):
new section 7.1.4 modifying relationship between gist/nslp for
handling route changes
route change now decomposed as add/delete rather than change
nslps can remember as many routes as they like and keep state on
them all
(was implicitly there previously to allow teardown on old path)

henning:
there is an issue of whether new path is temporary (routing protocol
delay) or a permanent alternative path

robert:
also relates to mobility draft - the authors of that should look at
this section and comment

version -13 to be released imminently

plan wglc to start 2nd April


QoS NSLP (Jukka)

various updates in v12

qspec stacking removed
initial version of abstract api

todo:
remove section 4.5

api:
probably needs to parts
rmf api
application api

dave oran:
this is one way to look at it
the direction is different for the two
appplication to nslp
nslp to rmf
key point is that there are two parts and the directions are different

i don't know if you want to suggest this:
tricky case is when pre-emption happens
in case when dealing with one reservation, rmf needs to talk to nslp
and application
probably would require protocol changes
i solved this for rsvp, but needed protocol machinery

christian:
is the application api also the api to a voip application?
jukka:
yes
in the diagram shown it is an 'internal' application


nat/fw nslp (Hannes):
only talking about most recent changes
various updates and clarifications

security area directorate review from catherine meadows

martin:
would be useful to provide initial review

lars:
i would recommend that WG does not do any more WGLCs while GIST is
still pending

Mobility draft (Takako)

chapters/sections restructured
main part explains how nsis protocols handle basic mobility mechanisms
further study part for enhancements
problem statement cleaned up

invalid nsis responder issue
how gist differentiates mobility event or dead peer

further work:
clean up security considerations
terminology
clean up further study section - currently still contains some
implementation matters

rob:
comment on invalid responder issue
always question in wireless access scenarios of how aggressive to be
i wonder if there is experience from other protocol designs for this
when i was talking to jukka, thought there may be something from dna

QSpec (Jerry):
main changes from ietf 67

recently mostly editorial updates in -15

before that significant changes in -13 and -14

WGLC
various issues arisen
QSpec Types
Initiator/Local QSPEC bit added
various editorial fixes

roland:
i've sent some comments
basically good shape
still some issues on what particular qos models can do with adding new
parameters and QSpec types

Y1541 went through WGLC a year ago
held pending stabilisation of qspec/qos nslp
set of updates now ready to submit

SCTP (Christian)
updates from Andrew McDonald's comments
removed TLS from draft

Andrew:
need a security solution in their
there is an individual draft on dtls over sctp, which looks a good
solution
however, it is an invidual draft, not sure of status

Magnus:
need a security mechanism

hannes:
agree with andrew
will check with chairs what status of this work is

Robert:
could remove pr-sctp, and just use tls with sctp

multistreaming:
only one sctp stream per sid
stream reuse for multiple sids is local policy

robert:
might use stream per priority
andrew:
though message ordering guarantees only guaranteed per stream

multi-homing:
might help smooth handling on route change

robert:
i need to read this section again
also worth looking at relevant part of gist spec again

i think this is irrelevant - MAs run between nodes, not along paths
can use multihoming to provide more failure resistant MAs
but this is orthogonal to path change

lars:
what benefit does sctp provide over tcp for gist?

christian:
not multihoming
maybe multistreaming to avoid head of line blocking

lars:
is head of line blocking a problem?

robert:
we can construct scenarios where it is a problem

robert:
multistreaming and partial reliability
no functional difference, just performance enhancement

it is similar issues to sip over sctp

dave:
no
sip proxy to sip proxy may have large amounts of traffic
also provide resistance to interface failure
gist doesn't want to have that resilience

rob:
the differences between tcp/sctp are in the performance
characteristics, as in the sip case

christian:
have implementation
roland:
so do we

martin: we are done