Multihoming - Resilience or Independence

That is very common procedure. In doing that at least 2 entries will be in
the
global routing table. And that's exactly the intension of the customer so in
case one path
failed(thus withdrwan properly ) the other path will kick in.

If the provider B lost both connections to the customer and provider A then
it lost the FULL
correct routing table then this is not the issue of filtering /20 anymore.
If provider B configure their BGP sessions right then it should withdraw all
the routes from provider A and
the customer. 

Ping Lu
Cable & Wireless USA
Network Tools and Analysis Group
W: +1-703-292-2359
E: plu at cw.net


> -----Original Message-----
> From: BLOCKI,JACEK (HP-Poland,ex1) [mailto:jacek_blocki at hp.com]
> Sent: Thursday, October 11, 2001 2:12 PM
> To: Lu, Ping; lir-wg at ripe.net; routing-wg at ripe.net
> Subject: RE: Multihoming - Resilience or Independence
> 
> 
> Hi,
> The customer will of course use 2 ISP - as requested. He will 
> use addresses
> coming from a single PA block advertised by two providers. Of 
> course you
> need to ensure 100% up connection between border routers. If 
> connection
> between border routers is down one of them has to stop 
> advertising PA block.
> Otherwise you have problem with return traffic. Imagine 
> provider B lost
> connection to both customer and provider A. If B is still 
> advertising common
> block some ASes may decide to reach customer over B, but B 
> cannot reach
> customer network... In order to deliver proposed service 
> providers A and B
> have to be connected with redundant network, which is quite common. 
> 
> Vicious circle (sp) as I see it:
> *Providers want small routing tables
> *Customers 100% up service so they as for resilience
> *Providers usually have single POP in area
> *Customers go BGP creating problem for community (table grows) and
> themselves (BGP expert costs)
> 
> Solution: 
> Providers enter into cooperation so they can advertise 
> customer addresses
> over 2 independent BGP speakers. This idea is not very different from
> existing aggregation. Specific routing information is hidden 
> in a group of
> ASes instead of single AS. Of course you need some 
> administrative effort
> (e.g. consistent billing) but from technical point of view it seems
> feasible. I may be wrong, it won't be for the first time ;-)
> Regards,
> Jacek
> 
> 
> -----Original Message-----
> From: Lu, Ping [mailto:PLu at cw.net]
> Sent: Thursday, October 11, 2001 5:02 PM
> To: 'BLOCKI,JACEK (HP-Poland,ex1)'; lir-wg at ripe.net; 
> routing-wg at ripe.net
> Subject: RE: Multihoming - Resilience or Independence
> 
> 
> 
> > -----Original Message-----
> > From: BLOCKI,JACEK (HP-Poland,ex1) [mailto:jacek_blocki at hp.com]
> > Sent: Wednesday, October 10, 2001 1:10 PM
> > To: lir-wg at ripe.net; routing-wg at ripe.net
> > Subject: RE: Multihoming - Resilience or Independence
> > 
> > 
> > Hi,
> > It seems multihoming discussion is an example of vicious 
> > circle (sp): More
> > specific routes result in larger routing table but we want 
> > those routes and
> > small table... Let me suggest a blasphemy: CIDR being 
> > advertised from two
> > ASes. They say it should be only one, but why? In my 
> opinion there is
> > nothing wrong in advertising CIDR from two ASes as long as 
> > you can guarantee
> > each border router advertises CIDR only if it can reach it. 
> > Imagine the
> > following construction:
> > AS-A--\
> >           +(OSPF) --(CIDR)
> > AS-B--/
> > 
> 
> That's the problem. When customer want to be muitl-homed you 
> can't tell them
> to use which ISP ? If the original prefix are not in the CIDR 
> with the other
> ISP
> then the customer have to change all their IP address. If 
> they don't then
> the second
> ISP have to leak the original prefix into global routing table.
> 
> The question is how do you convince your customer to change 
> all their IP
> addresses to
> have a second link without the risk of losing its business ? 
> 
> 
> > Each BGP speaker advertises CIDR if and only if it learned 
> > about it from
> > OSPF. It can be done, if you don't know how I'll forward 
> you a working
> > example. Each border router generates a default router and 
> > injects it into
> > OSPF. From technical point of view I see no reasons why it 
> > should not work.
> > What you need is:
> > * An agreement between ISPs
> > * Change in procedure making such a union of ASes an 
> > officially blessed
> > solution so nobody would dare to hinder cooperation with filters.
> > * Optionally you may need a separate CIDR, since both ASes 
> > have to advertise
> > same prefix. You need it if each ISP wants to have private 
> > customers in
> > addition to shared ones.
> > 
> > The customer has "an independent connection to two ISPs" 
> > which is the quest
> > item. I see more commercial than technical problems with such 
> > a solution.
> > However my expertise is limited and somebody can point 
> > drawbacks I cannot
> > see. Feel free to burn me on a stake, that's the right way of 
> > treating a
> > blasphemy ;-)
> > Regards,
> > Jacek
> >
> 
> It is a good idea but not easy to implement under the 
> pressure of making
> revenue.
> 
> 
> Ping Lu
> Cable & Wireless USA
> Network Tools and Analysis Group
> W: +1-703-292-2359
> E: plu at cw.net
>