The MPLS WG Archive

["Vijayanand C - CTD, Chennai." <vijayc@ctd.hcltech.com>]

Hello all,
I have an alternate suggestion for gracefully restarting LDP.I will just
briefly outline it below. I don't know if this approach has any flaws.

This method involves modifying the existing Hello and Keep alive
negotiations. Instead of negotiating them and maintaining same keep alive
value for a session and the same hello value on both sides we need to
maintain different timer values on both sides. This is not a major change to
the existing protocol. If this is done then either side can dynamically
change their Hello and keepalive values even after session initialization.
Hence if LDP is going down gracefully it can increase its Hello and Keep
alive values and advertise them before going down. Also the LDP going down
can store its TCP-UDP socket info and its session/adjacency info instead of
closing them. So the restart will be transparent to the peers. LDP may also
override TCP keep alive information. This works fine when LDP alone is going
down.  However in addition to LDP if TCP/UDP are also going down then they
should store their state info. and go down without releasing the resources,
so that peer TCP/UDPs dont' close the sessions.

This is only a brief outline and a few more things need to be done to
overcome other problems.

I am not sure if this approach is worthwhile.Please correct me if I am wrong
with your valuable feedbacks.

Regards,
Vijay