Re: [Simple] Other proposal from View Sharing and Common Notify to reduce presence traffic

[Victoria Beltrán Martínez <[e-mail]>]

Hi,

Certainly I considered a mechanism to exchange the policy documents of 
watched presentities between watching and watched domains in Common 
Subscribe solution. The watching domain (RLS) subscribes to an event 
(called “privacy filters”) in presentity domain’s presence server. 
Whenever a change in the policy documents for any of the presentity’s 
watchers happens, a NOTIFY message with “event=privacy filters” is sent 
from presentity’s presence server to watchers’ domain. The watching RLS 
maintains one subscription to “privacy filters” event per watched 
presentity. This subscription could be explicit if existing watchers 
were indicated into the body of SUBSCRIBE messages. In this way, when a 
new watcher is added a re-SUBSCRIBE message containing the updated 
watcher list is sent and a NOTIFY message with the policy documents will 
be received by the RLS in response to this resubscription. However, a 
privacy filters subscription could be implicit so that the watcher list 
is not needed in SUBSCRIBE messages, a simple SUBSCRIBE creates a 
subscription to all the existing watchers of the watched presentity. 
Clearly, a trust relationship between watching RLS and watched PS is 
needed to share full-state presence and policy documents of watched 
presentities.

When the watching domain subscribes to a presentity, the SUBSCRIBE 
message contains the list of existing watchers in order to let the 
watched domain know what watchers are watching the presentity. The OK 
response from the watched domain includes the list of authorized 
watchers to see some view of the presentity’s presence. A view of 
presence is defined by a policy document.

Assuming N watched presentities , with Common Subscribe the watching RLS 
will maintain 2*N subscriptions (N for presence event and N for privacy 
filters event). In case of View Sharing, the RLS will maintain N*F 
setting F as the number of different policy documents per presentity. At 
first sight, we can imagine how the performance of VS gets worse as F 
increases.

In addition, the fact that policy documents change rarely allows us one 
interesting possibility with Common Subscribe. When a common 
subscription is created for a presentity in the watched domain’s 
Presence Server, automatically a “logical” subscription to the policy 
documents associated to the presentity is created too. In this way, the 
watching domain receives two initial NOTIFY messages (or a single 
multipart NOTIFY), one contains the presentity’s presence and another 
one contains the policy documents. Once the policy documents are 
downloaded is more than likely that privacy rules don’t change during 
the session, so that we save unnecessary signaling traffic to keep alive 
subscriptions to privacy filters events. Anyway, if a change of privacy 
rules associated to a presentity happened it would be notified as a 
NOTIFY message into the common subscription between the presentity and 
the affected watchers’ domain. Now, if there are N presentities, the 
total number of subscriptions maintained by the watching RLS is N. I 
have called this mechanism as internal privacy filters subscription.

I did several assumptions to estimate the performance of Common 
Subscribe and View Sharing. I supposed the presence federation scenarios 
described in section 2.8 of the draft “Presence Interdomain Scaling 
Analysis for SIP/SIMPLE”) (ietf-simple-interdomain-scaling-analysis-04). 
I considered both implicit and internal privacy filters subscriptions 
when Common Subscribe was applied. With View Sharing, I agree with David 
that filter-based subscriptions are subject to less frequent change, so 
I supposed that each change affected half of the privacy filters on 
average (a notification was generated in half of the existing 
filter-based subscriptions on average). The number of filter-based 
subscriptions between watching and watched domains is equal to the 
number of different policy documents associated to watchers in the 
watching domain. I considered several numbers of policy documents 
depending on the number of watchers per presentity. My calculations show 
that Common Subscribe is more scalable than View Sharing and it achieves 
around 50% improvement of presence traffic in the majority of cases.

Regards,

Victoria


En/na David Viamonte ha escrit:
> Hi,
>
> I agree with Sanjay that, at first sight, it seems that for this 
> alternate proposal to work, a sharing of (part?) of the Presence 
> authorization policies between serving and watching domains is needed. 
> On the other hand, making a single "domain-to-domain" subscription as 
> Victoria suggests would be the simplest and probably most simple 
> mechanism to distribute Presence information across domains. In such 
> case the watching domain (RLS) would have responsibility of 
> implementing Presence policies about Presentities in remote domains.
>
> In practice, the need for sharing policy documents is implicitly 
> mentioned in the "Security Considerations" section of 
> "draft-ietf-sipping-presence-scaling-requirements-01". According to 
> that section, no decision has been made yet about allowing or 
> discouraging such approach (delegating privacy protection from one 
> domain to the other). On the contrary, it seems to be left FFS to 
> decide whether this is an acceptable security risk or not ("important 
> part of work on the requirements and optimizations will be to make 
> sure that all the security aspects are covered").
>
> Furthermore, my understanding is that, in practice, this 
> "domain-to-domain single subscription" mechanism does not require the 
> complete Presence or Policy documents being exchanged, only those that 
> might be applicable to any user in the watching domain (of course, 
> including a mechanism to update policies, if needed). In fact, I think 
> EXACTLY the same comments in the Security Considerations section of 
> the View Sharing draft ("if it generated a subscription from each of 
> its subscribers it would be able to determine who from its domain is 
> allowed to subscribe and what view they would receive") are applicable 
> to Victoria's proposal, so I fail to see substantial security 
> differences between both approaches. I could envisage potential 
> "logistic" issues, though...
>
> Since I am not a security expert, my conclusion is that getting an 
> understanding on whether there are fundamental diferences (from the 
> security perspective) between both approaches would be very valuable. 
> In case there are not such differences, the proposed alternative looks 
> interesting, as it seems to represent a very important simplification 
> (in practice, there would be no need to define any SIP extension at 
> all). It is fair to say that a mechanism to exchange policies would 
> need to be defined (not sure is that is covered elsewhere?).
>
> When it comes to traffic reduction, my understanding is that the 
> "common subscription" mechanism "should" be more efficient than the 
> view sharing approach. Even though it leads to larger documents being 
> exchanged (thus containing more Presence information, which is subject 
> to more frequent change), such documents are delivered only once per 
> domain, while in the view sharing case documents are delivered to each 
> list of watchers sharing a same "policy profile". This is only a 
> preliminary comment without having made serious calculations :-)
>
> Just my 2 cents,
>
> David
>
>
>
>
> Sanjay Sinha (sanjsinh) escribió:
>> What about allowed/blocked list that a presentity may have, if there is a single subscription between watcher domain's presence server and the watcher presentity.
>>
>>
>>
>>   
>>> -----Original Message-----
>>> From: [e-mail] [mailto:[e-mail]] 
>>> On Behalf Of Victoria Beltrán Martínez
>>> Sent: Wednesday, October 15, 2008 1:11 PM
>>> To: [e-mail]
>>> Subject: [Simple] Other proposal from View Sharing and Common 
>>> Notify to reduce presence traffic
>>>
>>> Hi,
>>>
>>> The Common Notify technique (described in the expired draft 
>>> "Scaling Optimizations for Presence in SIP/SIMPLE,
>>> houri-simple-interdomain-scaling-optimizations-00) is intended 
>>> for reducing presence traffic between federated domains. Other 
>>> draft, "Optimizing Federated Presence with View Sharing" 
>>> (ietf-simple-view-sharing-01) has the same intention. From 
>>> these proposals,  I was thinking the idea of a single 
>>> subscription between watcher and watched domains, so that the 
>>> presence server in the watcher domain is in charge of 
>>> subscribing to the watched presentity.  Then, the presence 
>>> server mantains the list of watchers that are watching each 
>>> presentity and this list hasn't to be sent in NOTIFY messages. 
>>> In this way, we avoid the increase of signaling traffic as the 
>>> number of watchers grows in the case of Common Notify and as 
>>> the number of privacy filters grows in the case of View Sharing.
>>>
>>> Have anyone considered this solution? I have read nothing about it.
>>>
>>> I named this solution as Common Subscribe and compared it with 
>>> Common Notify and View Sharing. I did several assumptions 
>>> based on IETF drafts and I proved that Common Subscribe 
>>> reduces around  50% of presence traffic generated by View 
>>> Sharing and Common Notify.
>>>
>>> thanks,
>>> Victoria
>>> _______________________________________________
>>> Simple mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/simple
>>>
>>>     
>> _______________________________________________
>> Simple mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/simple
>>   
>> ------------------------------------------------------------------------
>>
>>
>> No virus found in this incoming message.
>> Checked by AVG - http://www.avg.com 
>> Version: 8.0.173 / Virus Database: 270.8.1/1728 - Release Date: 16/10/2008 7:38
>>
>>   
> ------------------------------------------------------------------------
>
>
> No virus found in this outgoing message.
> Checked by AVG - http://www.avg.com 
> Version: 8.0.173 / Virus Database: 270.8.1/1728 - Release Date: 16/10/2008 7:38
>
>   

_______________________________________________
Simple mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/simple