[j-nsp] BGP Hold-times

[Jee Kay]

Hi all,

We're running BGP in an HA environment, using a hold-time of 3
seconds. This translates to a KA interval of 1s and is working for us
on all of our Ciscos.

The lowest hold-time you can confirm on the Junipers seems to be 9
seconds.. however as per the BGP spec if two devices disagree on the
timer then the lowest should be used. To me, this means that 3 seconds
should end up being used.

However, when the peering comes up the Juniper logs the following:

rpd[2701]: %DAEMON-4: bgp_get_open: peer x.y.28.229 (Internal AS z):
hold time too small (3), set to minimum (20)


But... a show neighbor shows the below:

Juniper> show bgp neighbor x.y.28.229
Peer: x.y.28.229+11038 AS z Local: x.y.129.218+179 AS z
  Holdtime: 9 Preference: 170


At this point I am confused.. the Cisco thinks the hold-time is 3
seconds. The Juniper seems to have disagreed and set it to 20 second
but the show neighbor says 9 seconds.

Has anyone seen this before? Is it possible to get the Junipers down
to a hold-time of 3?

Failing this, how are people handling sub-10s convergence when running BGP?

Thanks,
Ras