Re: [CCAMP] The description of Path Key retaining time in RFC5553

["Zhangxian (Xian)" <[e-mail]>]

Hi, Zafar, 

    Let me summarize a bit, so that whoever follow our discussion are not lost. 

1st viewpoint: you still think that 16-bit PKS is not sufficient, since it limits the number of a LSP within a network to be 64k. 
Our Reply: the purpose of this draft is to address the LSPs that across UNI and/or multi-domains, *NOT* the LSP within a network. Let's not argue theoretically it can be any number for the former. From current deployment, it is far from enough. Besides, we also provide a PKS+ first node of the encode segment as an alternative for the problem you worry much about, is it enough? 

2nd viewpoint: you believe that the PCE needs to store the PKS information longer than required as RFC5520 and you would like to give a name to it.
Our reply: We get your point and will add into our draft this information, as you suggested. However, we disagree to use the term "stateful" since people already accept the definition of stateful PCE in PCE WG. Also, it is meaningless to argue on the terminologies as long as we make the point clear in the draft. Please do not forget that the border node can be the "PCE" for decoding the PK (which you should be much aware of), storing such information is not an problem. Having said that, our draft do not limit the model of how PKS is decoded.

  As usual, we can agree to disagree, but I think we both have made the point clear enough till now. 

  Please do not forget the reason we write this draft. As you John already mentioned, we do not see any description in draft-ietf-ccamp-lsp-diversity about how the border node can decode the sub-object defined upon receiving the path diversity constraint, thus there is NO complete solution in your draft. I vaguely remember you mentioned a proprietary method for doing so, do you mean that whoever implement draft-ietf-ccamp-lsp-diversity are forced to need to consult the draft authors for a complete solution? Please consider this in the next update of this draft.


Cheers,
Xian(on behalf of all co-authors of the draft draft-zhang-ccamp-route-exclusion-pathkey)

-----Original Message-----
From: Zafar Ali (zali) [mailto:[e-mail]] 
Sent: 2013年11月16日 2:56
To: Zhangxian (Xian); Fatai Zhang
Cc: [e-mail]
Subject: Re: The description of Path Key retaining time in RFC5553

Hi Xian:

Please see in-line.

Thanks

Regards … Zafar


-----Original Message-----
From: "Zhangxian   (Xian)" <[e-mail]>
Date: Thursday, November 14, 2013 10:43 PM
To: zali <[e-mail]>, Fatai Zhang <[e-mail]>
Cc: "[e-mail]" <[e-mail]>
Subject: RE: The description of Path Key retaining time in RFC5553

>Hi, Zafar,
>   
>   Thank you for sharing your thought. But I disagree. Why having a
>16-bit PK space will constrain the network to have only 64k LSPs? PK is
>needed only when the LSP is needed to use as a constraint for path
>diversity.

I disagree - 

- How would PCC or PCE know ahead of time which LSP will be used for
diversity by another LSP in a future time? E.g., if you are signaling LSP1
at t1. How would PCC or PCE will know that some other LSP (LSP2) will be
requesting diversity from LSP1 at a future time t2.

> 
>
>   Furthermore, I do not see why storing PKS in PCE for a longer time,
>even if the lifetime of LSP, will cause any scalability issue. PKS can be
>used locally by the PCE (stateless) per node basis

Please stop calling a solution that mandate (path key, path) "states" to
be stored for lifetime of the connection as a stateless solution. It is
stateful w.r.t. Path info states. Let's define a new term for such
stateful PCE and use it. Furthermore, as you solution required PCE to keep
path states for life time of the connection, why not just use stateful
PCE, instead? 


>, i.e. combining PKS + source node address as a way for PCE to solve the
>issue you mentioned below, which is internal to the PCE.
>Do you agree?

No... In your solution a PCE cannot serve more than 64 paths in the
network. If you have a centralized PCE, you cannot have more than 64K LSPs
in the *entire network*. This does not fly.


> If needed, we can capture this in the manageability section.
>
>Regards,
>Xian
>
>-----Original Message-----
>From: Zafar Ali (zali) [mailto:[e-mail]]
>Sent: 2013年11月15日 11:13
>To: Fatai Zhang; Zhangxian (Xian)
>Cc: [e-mail]
>Subject: Re: The description of Path Key retaining time in RFC5553
>
>
>-----Original Message-----
>From: Fatai Zhang <[e-mail]>
>Date: Monday, November 11, 2013 9:09 PM
>To: zali <[e-mail]>, "Zhangxian (Xian)" <[e-mail]>
>Cc: "[e-mail]" <[e-mail]>
>Subject: RE: The description of Path Key retaining time in RFC5553
>
>>Please don't argue scaliblity issue because you can see even a GMPLS node
>>can store lots of informaiton.
>>
>
>Fatai- 
>
>I did not get what you are saying. What I was saying is:
>
>* Tunnel-id is 16 bit. I.e., client name space is 16 bit. There can be
>more than one LSP per tunnel.
>* Path key is 16 bits. I.e., server name space is 16 bits. Each LSP needs
>a Path Key. 
>* There is 1:N relationship between PCC and PCE (server). PCE sever may
>even be centralized.
>* You are requiring (Path key, path info) "state" to be stored at the PCE
>server for the life time of the connections.
>* Path keys have hold-off timer of 30 minutes before they can be reused.
>
>Hence, in your solution, PCE server would run out of Path Keys much before
>clients runs out of tunnel name space. In your solution, if you have a
>centralized PCE, you can only support 64K connection IN THE NETWORK! I am
>not sure why you want to ignore these scaling restrictions.
>
>Thanks
>
>Regards...Zafar 
>
>>
>