Superfluous advertisement (was: Friday's Random Comment)
A use case for a longer prefix with the same nexthop: F / \ D E | | B C \ / A Suppose A is a customer of B and C. B has a large address space: 10.1.0.0/16. B allocates a subset to A: 10.1.1.0/24. B advertises the longer prefix to its backup provider C. C propagates it to E and then to F. B MUST advertise both 10.1.0.0/16 and 10.1.1.0/24 to D. D MUST propagate both of them to F. Otherwise, if F only receives 10.1.0.0/16 from D, then F will have the longer match 10.1.1.0/24 to E, but E is only the backup route. Thanks, Jakob.
-----Original Message----- Date: Fri, 29 Apr 2016 08:17:41 -0400 From: Alain Hebert <ahebert@pubnix.net> To: "'NANOG list'" <nanog@nanog.org> Subject: Friday's Random Comment - About: Arista and FIB/RIB's Message-ID: <00ea292f-e779-25ad-ce89-eae897e9516d@pubnix.net> Content-Type: text/plain; charset=utf-8
While following that Arista chat... That reminded me of that little afternoon project years ago.
So I decided to find new hamsters, fire up that VM, refresh the DB's and from the view point of a tiny 7206VXR/G1 with 2 T3 peers...
The amount of superfluous subnet advertisement drop to ~120k from ~166k from the previous snapshot.
And this is the distribution by country.
country | superfluous --------------------+------------- United States | 28254 Brazil | 10012 China | 7537 India | 6449 Russian Federation | 4524 Korea, Republic of | 4062 Saudi Arabia | 3297 Australia | 2989 Indonesia | 2878 Hong Kong | 2251 Thailand | 2093 Canada | 2019 Taiwan | 1955 Ukraine | 1877 Singapore | 1856 Bulgaria | 1488 Argentina | 1436 Japan | 1403 Mexico | 1351 Chile | 1271
(Damn Canada, can't break top 10 again).
PS: "Superfluous" is a nice way to say that the best path of a subnet is the same as his supernet. And yes I'm aware of the Weekly Routing Report, I was just curious to see it by country =D.
----- Alain Hebert ahebert@pubnix.net PubNIX Inc. 50 boul. St-Charles P.O. Box 26770 Beaconsfield, Quebec H9W 6G7 Tel: 514-990-5911 http://www.pubnix.net Fax: 514-990-9443
A use case for a longer prefix with the same nexthop:
F / \ D E | | B C \ / A
Suppose A is a customer of B and C.
This is possible, but only remotely probable. In the real world, D and E are likely peered, as are B and C. Further, it's quite possible for F to choose the path through E anyway, regardless of A's wishes, or even to load share over to the two paths. If it's really a backup path, and you don't want traffic on it unless the primary is completely down, then you need to not advertise it until you actually need it. One of the various principles of packet based routing is that if you advertise reachability, it means someone, someplace, might just choose the path you've advertised. You can't control what other people choose. :-) Russ
Simpler, with B and C peered: F / \ B---C \ / A If B does not send the /24 to F, then F will send all the traffic to C, even if A wanted a load balance. Maybe I could ask the community: Why do you advertise longer prefixes with the same nexthop as the shoter prefix? Is it this use case, or something else? Thanks, Jakob.
-----Original Message----- From: Russ White [mailto:7riw77@gmail.com] Sent: Saturday, April 30, 2016 12:35 PM To: Jakob Heitz (jheitz) <jheitz@cisco.com>; nanog@nanog.org Subject: RE: Superfluous advertisement (was: Friday's Random Comment)
A use case for a longer prefix with the same nexthop:
F / \ D E | | B C \ / A
Suppose A is a customer of B and C.
This is possible, but only remotely probable. In the real world, D and E are likely peered, as are B and C. Further, it's quite possible for F to choose the path through E anyway, regardless of A's wishes, or even to load share over to the two paths. If it's really a backup path, and you don't want traffic on it unless the primary is completely down, then you need to not advertise it until you actually need it. One of the various principles of packet based routing is that if you advertise reachability, it means someone, someplace, might just choose the path you've advertised. You can't control what other people choose.
:-)
Russ
If B does not send the /24 to F, then F will send all the traffic to C, even if A wanted a load balance.
Maybe I could ask the community: Why do you advertise longer prefixes with the same nexthop as the shoter prefix? Is it this use case, or something else?
it is a common TE use case. but folk watching the water rise are starting to ask why the whole world should pay for A's TE. randy
F / \ D E | | B C \ / A
Suppose A is a customer of B and C.
This is possible, but only remotely probable. In the real world, D and E are likely peered, as are B and C.
"likely?" with what probability? any measurement cite please. nothing exact; something rough would be fine. randy
F / \ D E | | B C \ / A
Suppose A is a customer of B and C.
This is possible, but only remotely probable. In the real world, D and E are likely peered, as are B and C.
"likely?" with what probability? any measurement cite please. nothing exact; something rough would be fine.
Well, the average AS Path length is something like 4, and according to the charts Geoff has presented here and there, the graph is becoming more dense, as most people interconnect. The odds of finding an end-to-end path (4 hops) on the global 'net where no-one is peered in the middle seems pretty unlikely to me. It's not impossible, but it does seem unlikely, just given the average AS Path length and the density of the graph. For example, I suppose you could make A/B/C part of the same network which is intentionally not peered, or B/C two regional providers who are not peered with one another. You could then make D/E IXPs who have no transit connectivity between them, and then make F a tier 1 provider... But this really seems unlikely to me. How would you string together 4 AS' in a row that have no connectivity to any transit AS, even regional, like this? Two hops I can see, four I have a hard time seeing.
it is a common TE use case. but folk watching the water rise are starting to ask why the whole world should pay for A's TE.
Precisely. Tragedy of the commons. To put it in other terms, removing information reduces optimization -- but if I can get optimization by making someone else pay for the information, then, well, why not? :-) Russ
F / \ D E | | B C \ / A
Suppose A is a customer of B and C.
This is possible, but only remotely probable. In the real world, D and E are likely peered, as are B and C.
"likely?" with what probability? any measurement cite please. nothing exact; something rough would be fine.
Well, the average AS Path length is something like 4, and according to the charts Geoff has presented here and there, the graph is becoming more dense, as most people interconnect. The odds of finding an end-to-end path (4 hops) on the global 'net where no-one is peered in the middle seems pretty unlikely to me. It's not impossible, but it does seem unlikely, just given the average AS Path length and the density of the graph. For example, I suppose you could make A/B/C part of the same network which is intentionally not peered, or B/C two regional providers who are not peered with one another. You could then make D/E IXPs who have no transit connectivity between them, and then make F a tier 1 provider... But this really seems unlikely to me. How would you string together 4 AS' in a row that have no connectivity to any transit AS, even regional, like this? Two hops I can see, four I have a hard time seeing.
i was hoping for measurements, not seems unlikely. as you know, i am sceptical about our internet topology intuitions and modeling given how good bgp is at hiding information and how poor our vantage points are. ripe atlas, caida, etc. give us some view, but views with inconsistencies and contradictions. we could write a paper on the hazards of as topology. oh, we did. :) randy
participants (4)
-
Jakob Heitz (jheitz)
-
Randy Bush
-
Russ White
-
Valdis.Kletnieks@vt.edu