[nsp] Typical Prefix filters - greater than /24's dropped?

[Michael Lyngbøl]

On 20.11.2003 22:29:36 +0000, Gert Doering wrote:
> hi,
> 
> On Thu, Nov 20, 2003 at 10:58:48AM -0000, Sam Stickland wrote:
> > I'm trying to get a feel for the sort of prefix filters that are typically
> > in use on the backbones and transit providers. Are prefixes longer than
> > /24's typically filtered out?
> 
> We do not accept anything more-specific than a /24.

That's the basic standard.

> I'd really appreciate if people would stick to aggregation: announce your
> network block in *one* piece.  Don't try to impose the costs for your
> broken business models ("this upstream must only be used for backup, so
> I'll announce more-specific over the other upstream") onto the whole
> world.

Prepending mixed with community support (if existing) from your upstream
provider is more doable.

Philip Smith, Cisco gave a nice presentation at NANOG in Salt Lake City
<http://www.nanog.org/mtg-0306/smith.html>

> *Everybody* pays for every additional prefix, and it sums up.  When we hit
> 200.000 routes (or so), the next round of "junk all your backbone ciscos
> because they can't take more than 256 Mb of RAM" is due, and then you'll
> *feel* the cost.  Stop it before that happens.

It's a lost cause.

Inspired by Verio's BGP filter policy[1] filtering a full view yelds
~107K accpted routes and ~20K rejected:

Peer               AS      InPkt     OutPkt    OutQ   Flaps Last Up/Dwn State|#Active/Received/Damped...
111.111.111.111  3292    4374165     115848       0       3      5w5d5h Establ
  inet.0: 106879/127113/143


/Michael

[1] http://info.us.bb.verio.net/routing.html#PeerFilter

-- 
Michael Lyngbøl -- michael at lyngbol dot dk
Network Architect, AS3292 TDC, IP·backbone