[c-nsp] BGP maximum-prefix on ASR9000s

Lukas Tribus lists at ltri.eu
Mon Jan 27 04:23:21 EST 2020


Hello,

On Mon, 27 Jan 2020 at 08:14, Mark Tinka <mark.tinka at seacom.mu> wrote:
> On 27/Jan/20 08:05, Hank Nussbacher wrote:
>
> > As many of us run full routing tables on our ASR9000s, we have just
> > found popping up in our logs:
> > gp[1058]: %ROUTING-BGP-5-MAXPFX : No. of IPv4 Unicast prefixes
> > received from xxx.xxx.220.91 has reached 786433, max 1048576
> > Reference:
> > https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/asr9k_r5-3/routing/configuration/guide/b_routing_cg53xasr9k/b_routing_cg53xasr9k_chapter_010.html
> >
> > The undefined default for maximum-prefix on ASR9000s (IOS-XR) is 1048576.
> > Recommendation: increase maximum-prefix to 1500000
>
> Known issue since IOS XR launched back in the day.
>
> For as far back as I can remember (probably 2010 or earlier), we always
> had the below line as standard configuration in all our IOS XR platforms
> for BGP sessions that did not require a prefix limit:
>
>        maximum-prefix 4294967295 75
>
> Doesn't affect only the ASR9000, but all IOS XR platforms.

For people running full tables with labels (BGP-LU or
Internet-in-a-VRF), it's probably a good time to start thinking about
their label consumption, if a label is allocated per-prefix (default
in Cisco land at least for MPLS VPNs).

Running out of label space (with is limited to 1M, you can't stuff
more in a 20-bit label) is gonna be bad experience. While with the
6500/7600 TCAM issue only those particular nodes were affected, this
is not a question of obsolete HW, SW or TCAM partitioning on a
particular node and will affect all vendors in a per-prefix label
allocation configuration.

We are running Internet-in-a-VRF on both IOS-XE and IOS-XR, in per-ce
(meaning per next-hop) label allocation mode. It was buggy initially
in IOS-XE, but after a few rounds of bug-fixing a few years ago things
are running smoothly now. I like the fact that I'm only doing one L3
lookup on the ingress-PE, which is why I avoid per-VRF label
allocation mode.


cheers,
lukas


More information about the cisco-nsp mailing list