I've found there are many providers that have completely disconnected
autonomous systems. For example Yipes (6517) uses L3 on the west coast
and Williams on the east coast.
66.7.129.0/24 is advertised under their AS through WCG and
209.213.209.0/24 is advertised under their AS through L3.
And the number of connected autonomous systems with de-aggregated
prefixes appears to be even more common than a disconnected AS.
It would seem that many (most?) network operators are just ignoring the
more vocal opinions on NANOG.
There is a significant difference between not aggregating, and using
discontiguous-AS EBGP.
As long as you are familiar with the pitfalls, there is nothing
inherently wrong with using a single AS in multiple locations, and
advertising discrete blocks of address space in each one. The best reason
to do this is for a network that you eventually plan to merge - it
eliminates issues of having to make major BGP configuration changes.
Of course, it required you to point default routes out your upstreams, as
you will not see the prefixes from one discontiguous island, in another,
thanks to BGP loop detection.
Several large access ISPs have run in the fashion for extended periods.
As far as aggregation - they are a couple reasons to not aggregate, but
the vast majority of it is sloth.
Finally, in regard to "vocal opinions on NANOG" - well, anyone who has
read NANOG for a while knows that vocal isn't always correct.
Nothing inherently wrong with it if you're paying for transit, but good
luck getting peering in multiple locations without presenting consistent
views.
Just making sure Ralph knows this, since I'm sure achieving 99% peering
by getting 10GE into NYIIX is the goal for his OC192 over 2600 network.
Date: Wed, 13 Nov 2002 14:28:07 -0600 (CST)
From: Daniel Golding
Of course, it required you to point default routes out your
upstreams, as you will not see the prefixes from one
discontiguous island, in another, thanks to BGP loop
detection.
As far as aggregation - they are a couple reasons to not aggregate, but
the vast majority of it is sloth.
[...]
I've never seen anyone here complain that Yipes de-aggregates
66.7.128.0/18 into /24's like 66.7.129.0/24. Until the bigger providers
change their ways why should someone like me (who has only chopped a /20
into /21-/23 with a covering /20) decide that doing a single aggregate /20
announcement is going to make a difference?
Of course, it required you to point default routes out your upstreams, as
you will not see the prefixes from one discontiguous island, in another,
thanks to BGP loop detection.
ouch. bad practice defaulting like that, however to static route your individual
blocks wouldnt be a problem
Several large access ISPs have run in the fashion for extended periods.
whats the opposite of autonomous? dependent? so AS becomes DS?
seriously tho, if an AS ceases to be autonomous then theres little point in
having them and you may as well do global routing on prefixes with a hop count
hmm can RIP handle 120000 routes?
As far as aggregation - they are a couple reasons to not aggregate, but
the vast majority of it is sloth.
like to meet C&W peering policy etc? the only valid reasons imho are traffic
engineering and customer multihoming
Finally, in regard to "vocal opinions on NANOG" - well, anyone who has
read NANOG for a while knows that vocal isn't always correct.
altho it gives an indication of best practice and therefore policy and like the
other thread on filtering RIR allocation boundaries, sure you can go and do your
own thing but dont complain when providers start filtering your routes and
ignoring your prefixes!
Trying to run OC192 over a 2600 router would make more business sense than
giving away 250mbps of free transit, which you claim to have done (on
isp-bandwidth) lately.
I don't know how much of it is ignorance, or resource constraints. I've
worked with companies that have used disconnected AS's because they couldn't
justify multiple AS's and they needed to multihome in multiple isolated
locations. I've also worked with companies that deliberately de-aggregate
prefixes due to link capacity issues. Network operators have to balance
building an ideal network to what equipment, links, IPs and AS's they can
justify. I personally rather see network operators de-aggregate and utilize
space efficiently as opposed to hoarding larger prefixes to make their
network appear like something it isn't.
> inherently wrong with using a single AS in multiple locations, and
> advertising discrete blocks of address space in each one. The best reason
> to do this is for a network that you eventually plan to merge - it
> eliminates issues of having to make major BGP configuration changes.
Nothing inherently wrong with it if you're paying for transit, but good
luck getting peering in multiple locations without presenting consistent
views.
No problem at all. Use a tunnel.
Going back to the original question:
(A) Is there a reason have disconnected ASs? Sure. Does it make more sense
than using multiple AS numbers? No.
(B) Is there a reason to deaggregate? Absolutely. The biggest being rather
bad internal allocations practiced by networks.
I suppose that depends on how many static routes you would need, and how
many routers you would have to touch.
If you have 10 sites like this, and add or remove several blocks every day
(an extreme, of course), then you could end up manipulating many statics
on numerous routers, which, aside from being a waste of engineer time, can
lead to fat-finger mistakes.
Since when did default routing become bad form, on a transit-buying
network?
> Just making sure Ralph knows this, since I'm sure achieving 99% peering
> by getting 10GE into NYIIX is the goal for his OC192 over 2600 network.
Trying to run OC192 over a 2600 router would make more business sense than
giving away 250mbps of free transit, which you claim to have done (on
isp-bandwidth) lately.
Please note the difference between "giving away" and "this guy doesn't
pay".
I've never seen anyone here complain that Yipes de-aggregates
66.7.128.0/18 into /24's like 66.7.129.0/24. Until the bigger providers
change their ways why should someone like me (who has only chopped a /20
into /21-/23 with a covering /20) decide that doing a single aggregate
/20 announcement is going to make a difference?
First of all I've never seen anyone point to Yipes as an example of how to
do anything correctly or successfully.
Secondly, if you REALLY need to do it, you can probably get away with it.
But if you have any kind of decency and a common transit provider,
announce the aggregate and the more specifics with no-export. It's not
required, but it's what nice/smart people do.
Just because someone else litters doesn't mean you should too.
I suppose that depends on how many static routes you would need, and how
many routers you would have to touch.
If you have 10 sites like this, and add or remove several blocks every day
(an extreme, of course), then you could end up manipulating many statics
on numerous routers, which, aside from being a waste of engineer time, can
lead to fat-finger mistakes.
this is a hack whichever way you look at it.. just that its better than a
default and acheives a result more like the contigous AS would have had than an
end user network.. hmm i wonder if this would work if you ibgp peer your
discontigous border routers and use a route-map to make sure the routes point at
your upsteam - would remove the statics and your manual engineering issues.
argh what am i saying.. now i'm promoting this setup!
Since when did default routing become bad form, on a transit-buying
network?
if you are a proper ISP with a full routing table you dont need a default and
having one merely sends junk to your upstream, i guess thats chargeable so maybe
they think its a good thing but it doesnt really fit with the various nanog
threads on tidying up bogon packets as they hop around the net.
Actually, most of the RBOC/ILEC's use completely seperate AS's. "FCC
Regulation" being a legitimate reason to request a whole bushel of AS's
from ARIN.
Try doing an ARIN whois on bellsouth, and you get...