Re: [pcp] WG Call for Adoption: Optimizing NAT and Firewall Keepalives Using Port Control Protocol (PCP)

[<[e-mail]>]

Hi Markus,

Please see inline.

Cheers,
Med


>-----Message d'origine-----
>De : [e-mail] [mailto:[e-mail]]
>Envoyé : mardi 6 août 2013 15:15
>À : BOUCADAIR Mohamed OLNC/OLN; [e-mail]; [e-mail]
>Objet : RE: [pcp] WG Call for Adoption: Optimizing NAT and Firewall
>Keepalives Using Port Control Protocol (PCP)
>
>Hi,
>
>[e-mail] wrote:
>>
>> I have one comment for the authors: consider adding some text to
>highlight
>> PCP benefits to reduce keepalive messages when deployed in managed
>> networks (i.e., both the underlying network and the service are managed
>by
>> the same administrative entity). A typical example is SIP-based
>deployments:
>> optimize the load on access service nodes (SBC, SBE, P-CSCF, etc.)
>because
>> the lifetime of the mapping is known to the SIP UA and as such there is
>no
>> need to issue frequent register messages to maintain the mapping alive,
>etc.
>>
>
>Thanks for the comment.
>
>I have a question, though: If the same entity operates the network and the
[Med] This is indeed the same entity but distinct operational teams. 

>service such as SIP, do they need an explicit protocol like PCP to control
>or learn about the mapping lifetimes, or can they just configure them?
[Med] Configure (dedicated) longer timer is indeed an approach but the use of PCP is superior since the user agent can discover the configured lifetime and therefore react accordingly. 

 I
>mean, since the service provider is able to configure the NAT/FW mapping
>lifetimes in the way they like, can't they just set the keep-alives in
>their SIP clients accordingly?
[Med] As I said above, this is doable but it is static and needs off-line coordination to tweak the midbox and the user agent. 


 As an example, the Firewall in a mobile
>network can be configured to set the lifetime for TCP connections destined
>to the SIP proxy (SBC, P-CSCF) to 1 hour, and via Device Management the
>operator could also turn off or set the IMS client keep-alives to match
>that.
[Med] Agree; that's doable. 

 PCP would not be needed for that particular case.
[Med] PCP can be avoided but still there is a need for coordination between distinct operational teams (those operating the network and those operating the service). If for any reason the timers is to be updater because of some observed failures (synchronization issues), having PCP is superior sine the timers is hinted by the user agent and the user agent is aware of the timer without requiring an additional management interface. Note, pcp is not supposed to be activated only for SIP but for other purposes: one single solution for a variety of usages. 

>
>It would seem to me that PCP would have more utility in "unmanaged" or
>"unrelated" cases where the client app basically has no a priori knowledge
>about the network, and neither does the application service provider. Or
>similarly, the NATs or Firewalls would have no idea what the TCP
>connections would be used for.
>
>Markus