Cullen Jennings | 1 Aug 2004 06:07
Picon
Favicon
Gravatar

Comments on dialog-package 04


I was always concerned about how this would work on a simple black phone. To
work, the dialog state need to indicate the phone has gone off hook and is
collecting digits even though it has not sent an INVITE yet. Likewise it
needs to indicate when it was still off hook even though it had received a
BYE. As a side note, unlike the state transition diagram in 3261, all phones
I have seen have a state machine that includes these states.

I like that the examples show the state being reported as a dialog with no
to tag before the INVITE is sent. I think there should be explicit text in
the document explaining this. Actually I think that a new state would be
more appropriate but I can accomplish what I want if there is no new state
and text explaining a dialog with no to tag. I'm still not sure how I am
supposed to deal with it in the hang up case. If Alice has camped on Bob's
phone. If Bob receives a BYE, but has not actually hung up and Alice sends
an INVITE, Bob will return busy on phones that only support one call. Again,
I think another state is the best way to deal with this but other ways are
possible. 

I find the terminated state underspecified. How long will a device continue
reporting that something is in the terminated state after it terminates. 0
seconds? This would violate the event subscription idea that state data is
idempotent on subscribes. Infinite seconds? this clearly won't work. The
draft should suggest something that does work not just leave it as a hanging
open detail left for implementers to figure out.

The idea of reporting SDP here scares the crap out of me. SDP has things
like SRTP keying material. Do you really need to do this? None of the
applications described in the beginning offered any motivation for this. It
seems that if we are going to do this, we need to report two SDPs - the SDP
(Continue reading)

Orit Levin | 1 Aug 2004 06:33
Picon
Favicon

RE: Comments on conference-package-05

Cullen,
I think that the first question is who the audiences of the Package are.

If the audience is the participants - there is no reason for providing
them with this level of granularity. Moreover, with the hierarchal
approach, you propose, the focus may need to artificially "aggregate"
the multiple signaling sessions' information in order to achieve a
certain privacy level.

On the other hand, for moderators (using CPCP, for example) the
requested information (and even more details) will be needed.

Applying it to our work, this is a matter of conceptual decision: do we
want separate packages for participants and moderators, or do we want to
have a single package for both audiences? This will obviously require
the focus to filter and adjust the information based on the role of the
recipient in the conference.

Orit.

-----Original Message-----
From: Cullen Jennings [mailto:fluffy <at> cisco.com] 
Sent: Saturday, July 31, 2004 11:28 AM
To: Alan Johnston; Jonathan Rosenberg; Henning Schulzrinne; Orit Levin;
sipping
Subject: Re: [Sipping] Comments on conference-package-05

Please, just send me the XML for what the conference state will look
like
when the conference is currently out dialing Alice's office phone and it
(Continue reading)

Cullen Jennings | 1 Aug 2004 19:26
Picon
Favicon
Gravatar

consent-framework and identity


The more I think about this, the more I like it.

For this to work, clearly we need usable identity. So assuming the identity
draft was used, I think it helps eliminate the signature stuff on the
permission object. If Alice sends a grant of permission as a body, the
identity draft will both assert it really is Alice and it will also
integrity protect all the bodies to that the grant can not be tampered with.
This can help remove the very painful signature problem on the permission.

The permission is an authorization but it needs to include hints for how
strong the authentication should be. Clearly if I am authorizing "* <at> *" I
don't care about how strong the the authorization is and don't want to cause
useless round trip traffic. On the other hand if I am authorizing a team
member to have a complete view of of everything my phone is doing, I need
something strong. 

One of my confusions when I first read this had to do with having times in
the permission to make them automatically expire. I think some more text
around this could help. This is a good idea because if one of your
permissions allowed you to be DOS attacked, it is unlikely you will be able
to remove this permission while you are under attack. Having it just
automatically expire might be a good thing.

In section 4, the paragraph starting "It is important" (2nd from last para
in section) is confusing. Not sure what you mean.

Cullen

_______________________________________________
(Continue reading)

Cullen Jennings | 1 Aug 2004 19:34
Picon
Favicon
Gravatar

No Comments on consent-reqs 00


I really like these.

_______________________________________________
Sipping mailing list  https://www1.ietf.org/mailman/listinfo/sipping
This list is for NEW development of the application of SIP
Use sip-implementors <at> cs.columbia.edu for questions on current sip
Use sip <at> ietf.org for new developments of core SIP

Cullen Jennings | 1 Aug 2004 19:59
Picon
Favicon
Gravatar

uri-list-conferencing and multiple-refer


When would I use these two? They seem to do the same thing. What does the
SDP in this list-conf INVITE mean?

My take of this is that the list-conf stuff has Alice sending a INVITE to
get Alice into the conference. At the same time, Alice is doing an implicit
refer with a list asking the conference server to send an INVITE to all
theses participants. Unfortunaly merging these two requests into one will
result in not being able to return independent responses (including
authorization requests) to them.

I have not thought much about this but I wonder if a better idea is sending
an INVITE to create and join the adhoc conf then sending a REFER with a list
of participants to bring into the conference.

_______________________________________________
Sipping mailing list  https://www1.ietf.org/mailman/listinfo/sipping
This list is for NEW development of the application of SIP
Use sip-implementors <at> cs.columbia.edu for questions on current sip
Use sip <at> ietf.org for new developments of core SIP

Cullen Jennings | 1 Aug 2004 20:05
Picon
Favicon
Gravatar

comments on uri-list 00


Might want mention what should be done if a lists contains more than one
copy of the same target (ie. reject the whole request)

As a meta comment on all the list stuff, I don't want to split this up into
so many RFC that you can implement RFC X that allows you to be an exploder
and not implement any of the consent mechanisms for it that are in some
other RFC. I do like the per application approach we have take but we might
need to bundle this all up at some point.

I do not understand why the content disposition is needed. (no problems with
it, just don't know why needed)

An example with a list that has external references would help.

Cullen

_______________________________________________
Sipping mailing list  https://www1.ietf.org/mailman/listinfo/sipping
This list is for NEW development of the application of SIP
Use sip-implementors <at> cs.columbia.edu for questions on current sip
Use sip <at> ietf.org for new developments of core SIP

Cullen Jennings | 1 Aug 2004 20:09
Picon
Favicon
Gravatar

Comment on uri-services 00


List integrity is important but equally important given how the lists are
used is knowing who they came from. Both of these are deal with using the
identity draft. The identity provides strong integrity across all the
bodies. 

Give the current state of S/MIME, it is more or less non existent and I
would not want to make any security we expect to work rely on it.

Cullen

_______________________________________________
Sipping mailing list  https://www1.ietf.org/mailman/listinfo/sipping
This list is for NEW development of the application of SIP
Use sip-implementors <at> cs.columbia.edu for questions on current sip
Use sip <at> ietf.org for new developments of core SIP

Cullen Jennings | 1 Aug 2004 20:25
Picon
Favicon
Gravatar

Comments on event-throttle 01


Not clear if the goal of this is to limit the number of  messages or limit
the bandwidth of messages. The merging of partial states to a full state may
make the problem worse not better.

What happens if the queued messages grows too large? Or a messages is
delayed too long? I suspect that just terminating the subscription may be
the best answer.

I think you should extend the use case example to show that if Lisa had 100
friends that all changed their status every 10 minutes and Lisa could only
accept 3 updates per minute what would happen. Effectively the first change
"reserves" a spot on the queue to Lisa but when this spot gets to the top of
the queue, it is the most recent state update that gets sent and all the
others get removed from the queue. Effectively Lisa will get an update from
each friend every 30 minutes and it will happen in a "fair" way.

I don't buy into the statistical multiplexer model - I think that if any
state is lost, the subscription should be terminated. This still allows a
more recent full to replace discard an earlier full or partial and for
partials to be merged. If you go with this, the throttle mechanism will
never deliver invalid state but it will only delay the state that is
delivered. 

Not sure I am making sense here but if not, we can hash it out in a hallway
next week. You may be saying the same thing as me and I just don't
understand the draft.

Cullen

(Continue reading)

Cullen Jennings | 1 Aug 2004 20:37
Picon
Favicon
Gravatar

Re: Comments on conference-package-05


Good point - I agree we need to figure out what this is for. Once we have
that, we can figure out what to do but I suspect that depending which way we
go, either a bunch of stuff needs to get removed or we need to add this
level of granularity.

I would expect that advanced roster lists could be generated from this for
all participants. I was also imaging it was the place you went to get the
correlating information between things like SSRC, media streams, session,
users of session, and conference and sidebar. Effectively it was the place
that provided the mapping between all names.

Cullen

On 7/31/04 9:33 PM, "Orit Levin" <oritl <at> microsoft.com> wrote:

> Cullen,
> I think that the first question is who the audiences of the Package are.
> 
> If the audience is the participants - there is no reason for providing
> them with this level of granularity. Moreover, with the hierarchal
> approach, you propose, the focus may need to artificially "aggregate"
> the multiple signaling sessions' information in order to achieve a
> certain privacy level.
> 
> On the other hand, for moderators (using CPCP, for example) the
> requested information (and even more details) will be needed.
> 
> Applying it to our work, this is a matter of conceptual decision: do we
> want separate packages for participants and moderators, or do we want to
(Continue reading)

Dean Willis | 1 Aug 2004 23:04
Favicon

Re: Clarifications on SIP URI-List-Index 01.txt

George Foti (QA/EMC) wrote:
> Hi,
> 
> I have a couple of clarifications on the subject  draft:
> 
> 1) The 2 MIME types, application/resource-list-indices & application/resource-lists-bitmaps seem
to contain the same information in different formats.
> 
> Is this true, or have I misunderstood the differences.

Yes, this is mostly true.

However, there is a big difference in peformance for sparse vs. dense 
selections.

The resource-list-indices format is better for sparse selections. For 
example, let's say we have a list with 10,000 entries, but want to place 
a call to only three parties. In this case, we send the three indices 
that indicate the relevant three parties.

The resource-list-bitmap approach requires at least one bit in the 
selection bitmap for every possible target on the list. Consequently, if 
the list has 10,000 entries, we'd still have to send 10,000 bits just to 
select the three parties in the example above.

Conversely, assume I want to talk to 48 parties from a list of 64. This 
would require transmitting 48 discrete index values with the 
resource-list-indices approach, and only 64 bits in resource-list-bitmap 
approach.

(Continue reading)


Gmane