Re: draft-ietf-ngtrans-ipv4survey-02.txt

[Mathew Richardson <mrr@nexthop.com>]

> Akira Kato <kato@wide.ad.jp> [Fri, Sep 20, 2002 at 01:15:32AM +0900]:
>
>>>>		BGP - worldwide uniqueness needed
>>>I just thought it was worth mentioning that BGP doesn't require
>>>a worldwide, unique ID.  Indeed, it is my understanding that
>>>even in actual operational deployment, rfc1918 addresses are
>>>often used for BGP IDs.
>
>>	does the above still hold for EBGP routers?
>
>Yes. It is an ID rather than an address. You can not establish an
>(E|I)BGP session between a pair of routers with the same ID. If you
>use BGP route reflectors, the ID should be unique in the AS, as well.
>If you use BGP route servers, the ID should be unique in the IX, as
>well.

<snip>

Agreed with all of the above.  I'd say that operationally, even without
route-reflectors, IDs should be unique within the AS.  (IBGP peering
sessions often take place over addresses that correspond to the BGP ID,
for example.)

>In a private peering, randomly choosen IDs from RFC1918 space rarely
>cause a collision. It is better to have deterministic ways to define
>the IDs in bigger IXes, however.

<snip>

Also agreed.

As a final point, the current BGP specification does require that
the BGP Identifier be set to an IP[v4] address assigned to the
BGP speaker; however, there are deployed implementations that
don't require this.

mrr