On Nov 8, 2:39pm jhma@mcvax.org wrote:
holdtime-and-a-bit seconds. [Aside: shouldn't the session be refused at startup if a mutually agreeable keepalive value can't be negotiated rather than being allowed to flap like this?]
Junipers seem to be slightly in the wrong here. A hold time of zero is acceptable, though unwise, IMHO. In this case no keepalives are sent, and the systems have to rely on other means (such as the TCP connection) to ensure they're both alive, but one could easily imagine a situation where the TCP connection remains intact after the BGP process has gone to lunch. Not good. In any case, the systems are required to agree on the lowest of the two proposed hold times, except to note that hold times of 1 or 2 seconds are not allowed. Hence, the Junipers should accept and use the proposed hold time of 0. -- Per