Re: About IS-IS as PE/CE routing protocol in BGP/MPLS/VPN

This is for your information.

Thanks & Regards
Sheng

----- Original Message ----- 
From: "stefano previdi" <[email protected]>
To: "shengcheng" <[email protected]>
Cc: <[email protected]>; <[email protected]>; <[email protected]>; <[email protected]>
Sent: Friday, May 30, 2003 4:30 PM
Subject: Re: About IS-IS as PE/CE routing protocol in BGP/MPLS/VPN


> At 05:58 AM 5/30/2003, shengcheng wrote:
> >Stefano:
> >           As you have said "So, the consequence of this is that you have routing inconsistency 
> >  between sites with all the issues that are related ti it." . What do you mean the "routing inconsistency?"
> 
> 
> whatever changes in one site will be unknown to the other 
> one. This may result in routing loops (e.g.: a prefix is 
> withdrawn from one site and advertised by another one).
> 
> sham-link is considered as a p2p link. If flooding is 
> stopped because there is no other connectivity than the 
> sham-link, then the flooding scheme is broken.
> 
> 
> >Yes the database sychronization has been destroied, but i still think the route is right in PE/CE of 
> >both side. 
> 
> 
> The PEs _may_ have correct information but not the CEs and 
> not all the other routers in both sites. CEs and sites will 
> only have local information and out-of-date information 
> about the other site.
> 
> 
> >          It is really a disaster for IS-IS to encapsulate its packet in IP format, which is the only way to 
> >flood the LSP accross the sham link in my opinion. 
> 
> 
> if you want to flood between sites you have to use IP. Note 
> that you will encap also CSNP/PSNP packets because of reliable 
> flooding (IP is just the encap).
> 
> 
> >I think your suggention about shutting down the sham 
> >link when detecting the backdoor fails is a good idea. Maybe it is the only method that deserve our 
> >further attention when we have to fufill the database synchronization of IS-IS as a link state protocol.
> 
> 
> detecting the failure of the backdoor is not trivial. the PE 
> doesn't know who is the backdoor. The only way to do so is to 
> run an additional SPF with following constraints:
> 
>         1- do not use the sham-link in the computation
>         2- check reachability of the other PE
> 
> if the other PE is found after this special SPF than it means 
> the backdoor works. If not it means that the sham-link is the 
> only way to reach the other PE. But you understand that this 
> means the PE has to run one more SPF at each network change.
> 
> So now the question is: what do we do if the backdoor link 
> fails ? If you just shutdown the sham-link, then you completely 
> lose connectivity between sites and it is not the scope of the 
> game... 
> 
> So the alternative is to _always_ redistribute sites routes 
> into BGP (even if you do flooding over sham-link) so when the 
> sham-link is shutdown you can rely on mp-bgp to propagate routes 
> between sites. This has the side effect of flooding twice the 
> routes...
> 
> s.
> 
> 
> 
> >Do you have other constructive suggestion?
> >
> >  
> >thanks&regards
> >Sheng
> >
> >----- Original Message ----- 
> >From: "stefano previdi" <[email protected]>
> >To: "shengcheng" <[email protected]>
> >Cc: <[email protected]>; <[email protected]>; <[email protected]>; <[email protected]>
> >Sent: Thursday, May 29, 2003 9:09 PM
> >Subject: Re: About IS-IS as PE/CE routing protocol in BGP/MPLS/VPN
> >
> >
> >> At 12:49 PM 5/29/2003, shengcheng wrote:
> >> >Stefano:
> >> >
> >> >I don't think the flooding is necessary when the backdoor link fail.
> >> >First, the creation of sham link is just to remove the negative affect by 
> >> >backdoor link. when backdoor link fail, sham link is meaningless in fact.
> >> >Second, even we doesn't flush the lsa soon after the backdoor link fail, the 
> >> >right result of IS-IS spf route computation will not be affected. 
> >> 
> >> 
> >> I don't agree. If the backdoor link fails, then the two sites cannot 
> >> receive the lsp's of the other side. So whatever happens in the one 
> >> site will be hidden to the other. However, the sham-link will still 
> >> claim connectivity between the two. 
> >> 
> >> So, the consequence of this is that you have routing inconsistency 
> >> between sites with all the issues that are related ti it.
> >> 
> >> This is the reason why ospf does flooding over sham-link as well.
> >> 
> >> 
> >> >With the age of the LSA of the other site's PE and CE in one side, 
> >> >the database of the virtual one area (i mean "virtual" is because of sham link) 
> >> >may be destroied, but the route table on PE and CE will be still right . 
> >> >
> >> >Flooding over sham-link with encapsulation of isis packets into ip almost
> >> >can't be accepted.
> >> 
> >> 
> >> I know... but you don't have the choice if you want to be sure routing 
> >> information is consistent across sites.
> >> 
> >> Link-state protocols require that everyone get flooded with all routing 
> >> information. If you claim connectivity (sham-link) then you have to be 
> >> sure flooding will ALWAYS happen everywhere... which is not the case if 
> >> the bakdoor link fails.
> >> 
> >> One solution is to shutdown the sham link if the backdoor fails but in 
> >> order for this to be possible you have to detect such failure... not 
> >> easy without running additional SPFs.
> >> 
> >> s.
> >> 
> >> 
> >> >expect your kindly further discussion..
> >> >
> >> >regards
> >> >- sheng
> >> >
> >> >----- Original Message ----- 
> >> >From: "stefano previdi" <[email protected]>
> >> >To: "shengcheng" <[email protected]>; <[email protected]>
> >> >Cc: <[email protected]>; <[email protected]>; <[email protected]>
> >> >Sent: Thursday, May 29, 2003 3:18 PM
> >> >Subject: Re: About IS-IS as PE/CE routing protocol in BGP/MPLS/VPN
> >> >
> >> >
> >> >On Thursday 29 May 2003 04:11, shengcheng wrote:
> >> >> 
> >> >>
> >> >> Dear Stefano and Manav :
> >> >>     Thanks for your attention to our draft of ISIS as PE/CE 
> >> >> routing protocol in BGP/MPLS VPN.      
> >> >>     Unluckily, my email box doesn't work normally recently. 
> >> >> This is why we give such a late response. 
> >> >>     Your issue is really very helpful. We are trying to 
> >> >> deliver next version of the draft to update the existing 
> >> >> problem(including your issues) in the current version this week. 
> >> >>     Kindly wish you continue to pay attention to the draft !
> >> >>
> >> >> thanks & regards
> >> >> Sheng
> >> >
> >> >there is something else that need to be worked out. If the 
> >> >backdoor link between two sites failed, then the two sites 
> >> >do not receive LSPs anymore but still believe they have 
> >> >full connectivity through the sham-link. 
> >> >
> >> >I'm affraid we will have to specify that flooding should 
> >> >occur over the sham link. This complexify a bit the 
> >> >architecture (not to mention that flooding over sham-link 
> >> >implies encapsulation of isis packets into ip).
> >> >
> >> >s.
> >> 
>