Only 1 question: What about the companies that have a /24 out of the /20 0r /21 that are multi-homed? If the route rules are not carefully prepared the multi-homed customer then might be single-homed and tied to the upstream they got the IP's from. Thoughts? Jim
-----Original Message----- From: bdragon@gweep.net [mailto:bdragon@gweep.net] Sent: Thursday, April 17, 2003 3:56 PM To: nanog@merit.edu Subject: selective auto-aggregation
I believe we're losing the aggregation war. More and more entities are deaggregating, and not announcing their largest aggregates which makes prefix-length filtering less effective.
I'ld like to propose the concept of selective aggregation, whereby a router can be configured to aggregate based upon rules.
For example, if an RIR allocation boundary for a particular /8 is /21, that routes which are longer than /21 could be aggregated to /21 rather than discarded. Obviously, this would only be effective facing one's transit, since aggregating a peer would violate most peering agreements. In transit-free networks, this functionality would not be useful.
Similarly, the ability to auto-aggregate contiguous networks originated by the same AS, which could be applied even to routes with lengths shorter than an RIR boundary. This functionality could be useful facing ones peers.
This type of thing would need to be selective, since permitted deaggregation (no-export tagged routes with meaningful MEDs) can still be useful between entities which agree to such things.
If we can, I'ld like to avoid a holy war on whether deaggregating is someone's god given right, and stick to the premise that there are networks who will enforce aggregation policies, and want to do so in the most effective manner possible.
McBurnett, Jim wrote:
Only 1 question: What about the companies that have a /24 out of the /20 0r /21 that are multi-homed? If the route rules are not carefully prepared the multi-homed customer then might be single-homed and tied to the upstream they got the IP's from.
Thoughts?
I was curious how much of the de-aggregation was due to multi-homed companies requiring the longer prefixes. Are there companies that actually announce their smaller routes despite controlling the shorter prefix? What would be the benefit of doing so? -Jack
I do it. Reasons are to control inbound traffic better, some routes are annnounced to all upstreams, some are to some specific ones, some have different communities tagged to them, etc. I have one block which is /19 which I do not announce and only announce multiple /23, /24 out of. It works just fine - I'v not seen one place which does not route there...
I was curious how much of the de-aggregation was due to multi-homed companies requiring the longer prefixes. Are there companies that actually announce their smaller routes despite controlling the shorter prefix? What would be the benefit of doing so?
-Jack
{Historical lessons of atomic aggregates and the dangers of passing them along should be background} On Thu, Apr 17, 2003 at 05:04:06PM -0500, Jack Bates wrote: [snip]
companies requiring the longer prefixes. Are there companies that actually announce their smaller routes despite controlling the shorter prefix?
Yes.
What would be the benefit of doing so?
They mistakenly believe that all providers will proagate their more-specifics and want to attranct traffic in a certain way for a certain longest-match. If they - anticipate this link-juggling to ONLY occur along contracted paths - appropriately tag NO-EXPORT - also announce the greater aggregate ...then they'll get what they want out of the parties with whom they contract. It is trivial and stunning that service providers don't actively promote it to their customers. Some would rather collect money for customers grazing on the commons rather than for providing *service*. -- RSUC / GweepNet / Spunk / FnB / Usenix / SAGE
Only 1 question: What about the companies that have a /24 out of the /20 0r /21 that are = multi-homed? If the route rules are not carefully prepared the multi-homed customer = then might be single-homed and tied to the upstream they got the IP's = from.
Thoughts?
Jim
I'm not seeing any scenarios which would differ from the existing practice of filtering out said /24 entirely. Depending on how a particular entity heard the /24, they may even continue to accept it as such without aggregation. For example, if the company was announcing the /24 to a peer of a network using this type of aggregation, the peer would still see the route as a /24. Otherwise, they lose nothing by aggregating to the /20 or /21 along their transit path.
participants (5)
-
bdragon＠gweep.net -
Jack Bates -
Joe Provo -
McBurnett, Jim -
william＠elan.net