Hi, 

   Please see inline: 
________________________________________
发件人: Giovanni Martinelli (giomarti) [[email protected]]
发送时间: 2013年8月1日 6:30
到: Zhangxian (Xian)
Cc: [email protected]
主题: Re: comment on draft-zhx-pce-stateful-lsp-sync-00

Hi Xian,

On Jul 31, 2013, at 23:14 , "Zhangxian (Xian)" <[email protected]>
 wrote:

> Hi,
>
>   Thank you for questions.
>
>  I do not understand your statement "if the PCC state is so big that could 
> not be transferred in one shot (not necessarily one packet or one datagram). 
> " When we are talking here is to synchronize the LSP-DB, which consists of 
> LSPs of manageable message size. Actually, it is possible to combine multiple 
> LSP states together.
>

GM> In  section 1.1 "Motivation" you provide some numbers. Could you estimate 
the the LSP-DB size & bandwidth available? Do you agree that optimisation 
suggested here has some costs?
Xian>   
"The bible" provides a mechanism for full syncrinization with LSP-DB (say  with 
the size of X), our proposal is to allow incremental syncronization in which 
case you transfer partial DB (let's assume Y). Note that Y is always smaller 
than X. So IMO there is no extra cost with respect to DB size or bandwidth 
availaible. Incremental sync. can actually help the PCE to come into full 
function quicker. Having said that, we are not trying to get rid of full sync., 
but rather to accommodate the scenario where there is such need (i.e., the 
scenario provided in motivation).   

> Secondly, for what you mentioned "first request fails", what do you mean by 
> that? Since PCEP is based on TCP, so there should be mechanisms to support 
> reliable delivery (outside of PCEP). So, NO MATTER if it is full sync. or 
> partial sync., the DB will be incomplete anyway even if your assumption 
> holds. So it does not invalidate our proposal.

GM> I guess I meant a state syncronization operation (or sequence ) as per 
section 5.4 "State Synchronisation"  of the bible.  In general operations may 
fails for different reasons ... anyway TCP should help you in avoiding further 
mechanism.
Xian> Ok. 

>
>
> As for the last comment, if you read (or may be interested to) the WG draft 
> carefully, the function you mentioned below as "A PCC SHOULD remember the 
> deleted LSP as well" is already supported there. So, I do not see how it is 
> an issue at all.

GM>  could you provide a pointer where is already supported?

Xian> Let me start again. As stated in WG draft "This is accomplished by 
keeping track
   of the changes to the LSP State Database, using a database version
   called the LSP State Database Version."  and support of R bit in LSP object 
in Section 7.3, sending the remove state via PCRpt to stateful PCE for deleted 
LSP is already a part of WG document. In this document this state needs to be 
remembered for a little while longer to facilitate incremental LSP sysnc. Note 
that this is not too much of an overhead as this needs to be remembered only if 
there is a PCEP session down with which the PCC can do incremental update. 


BR,
Xian

Cheers
G


>
> Regards,
> Xian
>
> ________________________________________
> 发件人: [email protected] [[email protected]] 代表 Giovanni Martinelli 
> (giomarti) [[email protected]]
> 发送时间: 2013年8月1日 1:50
> 到: [email protected]
> 主题: [Pce] comment on draft-zhx-pce-stateful-lsp-sync-00
>
> Dear Authors,
>
> here an engineer (not necessarily a network engineer) comment. It comes to my 
> mind during today presentation so ...  tossing to the list hope help 
> discussion.
>
> Looks to me you are adding an optimisation that actually may add more 
> problems than the one you are trying to solve.
>
> I first wonder if the PCC state is so big that could not be transferred in 
> one shot (not necessarily one packet or one datagram).  In case first request 
> fails wound't be better just retry instead of adding the complexity of 
> incremental updates?
>
> Second, you have to pay the price of partially synchronised DB. What you can 
> do with a partial state? My first feeling you have to just consider your  DB 
> as not synchronised.
>
> Third, You also imply there's more state to maintain at the PCC side as 
> well...  e.g. "A PCC SHOULD remember the deleted LSP as well" .
>
> Cheers
> G
>
>
>
>
> _______________________________________________
> Pce mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/pce
_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce

Reply via email to