At 3:46 AM +0000 8/4/04, Peter van Oene wrote:
>At 12:49 PM 8/3/2004, neteng wrote:
>>I talked with some others that know BGP better than I do and they thought
it
>>might be achievable using MED. Any ideas on this from the BGP experts?
>>
>> The multi-exit discriminator (MED) attribute is a hint to external
>>neighbors about the preferred path into an AS when there are multiple entry
>>points into the AS. A lower MED value is preferred over a higher MED value.
>>The default value of the MED attribute is 0.
>
>It is equally common, or more so, and safer in my opinion, to use
>communities set by customers to influence the local pref used by the
provider.
I agree completely, Peter, that communities are by far the best way
to do this. They are easier to maintain, and if the SP decides to
change the way it influences its BGP selection algorithm, the
mechanics of that change are hidden from customers.
As you well know, however, the presence of a mechanism doesn't mean a
service provider is willing at all to implement a policy variation,
or will do it only for special fees. Any time a customer requests
something that might steer traffic away from the ISP, and thus reduce
its potential revenue, the customer had better have a good financial
argument ready on why the carrier should do what is wanted.
This layer 8 or so negotiation is apt to be more challenging than the
actual routing protocol. From a pure routing standpoint, it also can
be quite challenging to define the semantics of the community. My
usual preference is to write the policy effects at least informally
in RPSL and check out the logic at that level of abstraction. I use
that method -- RPSL before BGP -- in my books.
Incidentally, there is a freeware tool, RtConfig, which will take
RPSL policy statements and generate the majority of IOS configuration
statements needed to implement it. There were variants for
Wellfleet/Bay/Nortel, GateD, and RSD as well, although the Cisco
version is best maintained. Is there a JunOS version of RtConfig? It
probably wouldn't be hugely different from the GateD version.
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=91486&t=91342
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
