I was in a discussion the other day and several Tier2 providers were talking about the idea of adjusting their BGP filters to accept prefixes smaller than a /24. A few were saying they thought about going down to as small as a /27. This was mainly due to more networks coming online and not having even a /24 of IPv4 space. The first argument is against this is the potential bloat the global routing table could have. Many folks have worked hard for years to summarize and such. others were saying they would do a /26 or bigger.
However, what do we do about the new networks which want to do BGP but only can get small allocations from someone (either a RIR or one of their upstreams)?
Just throwing that out there. Seems like an interesting discussion.
Justin Wilson
j2sw@mtin.net
Hi,
this would at least help to get rid of many old routing engines around the world 
... or people would keep their "learn nothing smaller than /24" filters in place. Also an option - but not for companies who act as an IP transit provider.
best regards
Jürgen Jaritsch
Head of Network & Infrastructure
ANEXIA Internetdienstleistungs GmbH
Telefon: +43-5-0556-300
Telefax: +43-5-0556-500
E-Mail: JJaritsch@anexia-it.com
Web: http://www.anexia-it.com
Anschrift Hauptsitz Klagenfurt: Feldkirchnerstraße 140, 9020 Klagenfurt
Geschäftsführer: Alexander Windbichler
Firmenbuch: FN 289918a | Gerichtsstand: Klagenfurt | UID-Nummer: AT U63216601
A /24 isn't that expensive yet...
Matthew Kaufman
Hi Justin,
Rent or sell them a /24 and make money. If they can't afford a /24 at
today's market rate, why should the rest of us spend much more money
upgrading routers to accommodate their advertisement?
The annual systemic cost of carrying that prefix is still more than
double the one-time cost of acquiring a /24. No doubt that gap will
close, but there's no cost justification to change the /24 filters
just yet.
Regards,
Bill Herrin
Much m ore than I'm willing to spend. 
Cheaper than buying everyone TCAM
Matthew Kaufman
How many routers out there have this limitation? A $100 router I bought ten years ago could manage many full tables. If someone's network can't match that today, should I really have any pity for them?
Besides which more than one provider filters by a minimum prefix length per /8 - wasn't Swisscom or someone similar doing that? So multi homing with even a /24 is somewhat patchy in terms of effectiveness
--srs
In my view, no. Hardware-based routers with sufficient RIB/FIB/TCAM are table-stakes for edge connectivity.
But it's easy for me to spend other people's money.
;>
Welcome to the real world ...
Cisco SUP720-3BXL
Cisco RSP720-3BXL
and even the new and shiny SUP2T only supports 1 Mio routes (dicvided to IPv4 MPLS, IPv4 VRF, IPv4 global routes, etc).
I guess this is still the truth: there are at least a few ten thousand of these devices running big parts of the internet. Take a look at some big players network - e.g. Level3. Their customer access routers in Slovakia, Austria and Germany are still based on the Cisco 6500/7600 platform.
Of course there are many other vendors and platforms available which do NOT have this limitations. But there are also at least a ton of vendors on the market with exactly the same limitation :(.
best regards
Jürgen Jaritsch
Head of Network & Infrastructure
ANEXIA Internetdienstleistungs GmbH
Telefon: +43-5-0556-300
Telefax: +43-5-0556-500
E-Mail: JJaritsch@anexia-it.com
Web: http://www.anexia-it.com
Anschrift Hauptsitz Klagenfurt: Feldkirchnerstraße 140, 9020 Klagenfurt
Geschäftsführer: Alexander Windbichler
Firmenbuch: FN 289918a | Gerichtsstand: Klagenfurt | UID-Nummer: AT U63216601
Any RIR - or LIR - that considers allocating space in sizes smaller than
a /24 (for the purpose of announcing to the DFZ) would do well to read
this report from RIPE Labs:
https://labs.ripe.net/Members/emileaben/has-the-routability-of-longer-than-24-prefixes-changed
tl;dr: it's still a bad idea to allocate smaller than a /24.
On top of this, I've recently seen some figures that put a 'regular' BGP
table mix, at over half of the prefixes received (from numerous
upstreams) as being /24s. I really don't want to see everyone already
de-aggregating their /18s to /24s, to then go and de-aggregate down to
/27s instead.
Whilst getting routers with *big RIBS* for little monies, is easy (i.e.
Linux box + Quagga). Getting routers that have all the features SPs
need, with the throughput requirements too, /and/ have plenty of *FIB*
space - that's expensive. Super expensive.
Hi Suresh,
That hasn't been true for something like a decade. Anybody who filters
anything shorter than /24 without also taking a default route (or the
equivalent) is not fully connected to the Internet.
Regards,
Bill Herrin
* tom@ninjabadger.net (Tom Hill) [Fri 02 Oct 2015, 18:34 CEST]:
Any RIR - or LIR - that considers allocating space in sizes smaller than a /24 (for the purpose of announcing to the DFZ) would do well to read this report from RIPE Labs:
https://labs.ripe.net/Members/emileaben/has-the-routability-of-longer-than-24-prefixes-changed
tl;dr: it's still a bad idea to allocate smaller than a /24.
RIPE has long allocated up to /29. Not everybody needs addresses for the Internet; some just need a guarantee of global uniqueness.
-- Niels.
Are you suggesting that the Tier 1 and 2's that I connect to are not
filtering out anything shorter than /24? My expectation is that they are
dropping shorter than /24, just like I am.
Correct me if I'm wrong, but every *NOG BGP best practices document I've
read has advocated dropping all prefixes shorter than /24 at ingress and
egress.
There would be a default route sure - but the filter simply means that if your packets from say a src IP in a level 3 /24 (where the minimum alloc size was what, /20) wouldn't go through if you sent them though say a cogent interface
--srs
filtering out anything shorter than /24? My expectation is that they are
dropping shorter than /24, just like I am.
You mean longer. A /16 is shorter than /24. A /28 is longer. More 1 bits in
a row.
-Bill
There are lots of transits that will take le 32 on their customers inbound
but filter le 24 on egress announcements.
My incorrect verbiage aside, what did you think about the question I asked?
In a message written on Fri, Oct 02, 2015 at 11:47:31AM -0500, Jason Baugher wrote:
Are you suggesting that the Tier 1 and 2's that I connect to are not
filtering out anything shorter than /24? My expectation is that they are
dropping shorter than /24, just like I am.
Not exactly, but it's not what the other poster is implying either.
Many providers let a customer multi-home to the provider. That is
they provide two circuits from two different POPs to the customer.
Allocate the customer a /27-/29 from the provider's supernet. The
customer announces these small blocks back to the provider to get
high availability. The provider does not announce externally, because
it is part of the supernet.
In Cisco speak:
ip prefix-list my-supernets-small-subnets permit 10.0.0.0/8 ge 24
ip prefix-list my-supernets-small-subnets permit 172.16.0.0/12 ge 24
!
...some route-map customer-in stuff...
!
route-map customer-in permit 100
match ip prefix-list my-supernets-small-subnets
set community 1234:1234 1234:5678 no-export
!
...some route-map customer-in stuff...
!
Yes, many tier 1's will allow longer than /24 _from their customers_
and _out of their supernets_, and will not reannounce them.
Bill, I see where I went wrong now that I went back and re-read your
comment. I was conflating "longer" and "shorter". Thanks for your patience
on this trying Friday.