RE: BGP in Confederations and Using Synchronization

From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Tue Apr 15 2003 - 18:35:29 GMT-3


At 1:32 PM -0400 4/15/03, OhioHondo wrote:
>Howard
>
>Thank you for your reply. Since your answer indicates a single IGP domain
>for a sub-AS or a non-confederated AS, I have a followup if I may. Assume
>OSPF as the IGP.
>
>If I have an AS or a sub-AS that consists of several routers running EBGP to
>sevaral external AS's and sub-AS's --- and running IBGP of course, what is a
>scenario where synchronization works without causing problems??

First, let me apologize for the answer that leaps to mind -- the main
purpose of synchronization is causing trouble! Synchronization has
long been superceded in operational practice as we gained better
understanding of BGP, but, unfortunately, it's something that Cisco
exams still love.

>
>Within a sub-AS or non-confederated AS you would run a full mesh or route
>reflection. In simple scenario's the concept of synchronization is easily
>explained, but when I lab it up in a larger environment -- synchronization
>(along with the only propagate the Best Route rule, interferes with the
>coordinated effort of the IBGP routers to establish a common policy.
>
>An example --- setting the Local Preference on routes received from external
>AS's or sub-AS's. If rtrA learns a route via EBGP and installs it in its' IP
>routing table, and rtrB learns the same route but gives it a higher Local
>Preference, the route from rtrB will not be synchronized at rtrA or any
>route-reflector clients of rtrA.
>
>I would consider this a problem. Is there a workaround???

Oh, it is a problem. See RFC 3345 for an extensive discussion of the
persistent oscillation condition. Bottom line -- best current
practice says you MUST adjust all route preferences, be they MEDs or
local prefs, or IGP metrics, so that you always prefer
intra-confederation or intra-cluster routes. This is done much for
the same reason you always prefer an OSPF intra-area route.

>
>I've read comments in books that state synchronization should not be used
>with full mesh or RR. (Note that if each sub-AS in a confederation has its'
>own IGP domain, there are no other options.) What's the story??? ;)

I'd have to ask if these were exam prep or BGP/ISP operations books.
There is no viable reason I can think of to use synchronization in a
real-world network.

>
>-----Original Message-----
>From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
>Howard C. Berkowitz
>Sent: Monday, April 14, 2003 10:28 PM
>To: Group Study
>Subject: Re: BGP in Confederations and Using Synchronization
>
>
>At 8:58 PM -0400 4/14/03, OhioHondo wrote:
>>Hello
>>>
>> I thought I'd try resending this because the orginal thread
>>got somewhat
>>off track. Someone suggested that, when using confederations, each sub-AS
>>should be its' own IGP domain (i.e. with OSPF, have it's own area 0) Does
>>everyone else agree with this??
>
>Emphatically agree. The whole point of confederations is to bound a
>community of interest, with the reduction of iBGP loads in comparison
>with non-confederation a desirable side effect. If you are trying to
>bound BGP information, what would be the point of leaking IGP
>information outside the community of interest?
>
>*sigh* I suppose I can't preclude something like this on a lab, given
>Cisco's propensity for wanting to explore knowledge of knobs with
>configurations which, if anybody did in the real world, should be
>grounds for firing. But from my perspective, it really makes sense
>to master the plausible configurations first, and learn the protocols
>in the way they were intended to be used. If you get sufficient
>understanding of this and then get a weird case in the lab, your
>in-depth understanding should dig you out. It's simply not possible
>to practice every conceivable permutation of protocol options,
>especially those that don't make sense.
>
>>
>>Thge original question was....
>>I have a question for someone with a grasp on BGP Confederations and how>
>>synchronization affects it. My scenario
>>>
>>> AS301/R7 --- AS501/sub-AS65001/R3 --- AS501/sub-AS65002/R1 --- AS101/R9
> >> |
>>> |
>>> AS501/subAS65001/R2
>>>
>>> Routers R1, R2 and R3 are in a single OSPF domain. The BGP router-id's
>and
>>> OSPF router-ids are the same on all routers.
>>>
>>> My problem --- an advertisement comes in from AS101/R9, let say
>>49.0.0.0/8.
>>> That advertisement is propagated via OSPF to router R2 with R1's OSPF
>>> router-id.
>>>
>>> When that advertisement crosses the sub-AS border between R1 and R3, the
>>BGP
>>> router-id is changed to that of R3, therefor when the iBGP route gets to
>>R2,
>>> the BGP router-id is from R3 while the OSPF router-id is from R1. The
>>result
>>> is no sync. Any advice????
>>
>>Jerry Haverkos
>>jhaverkos@columbus.rr.com
>>614-351-8617



This archive was generated by hypermail 2.1.4 : Thu May 01 2003 - 13:35:53 GMT-3