Service Parameters, Callmanager, "Clusterwide Parameters (Device - SIP)". I tweak INVITE retries, for example down to 2 which mean three tries in total. The relevant timer is "SIP Trying Timer" which determines how long CUCM waits to receive a Trying response after sending an Invite. Bear in mind a couple of things. Firstly the timer is doubled on each retry, so for example at the default 500ms it will retry after 500ms, then next retry at 1 second, then 2 seconds etc. The other thing is that if you're using TCP it doesn't actually retransmit if the initial Invite went OK at the TCP level, it just waits the equivalent time.
If you set these timers and retries to suit, you should get a system where if the gateway goes down you only add a couple of seconds delay to outbound calls for the first minute or so, then once the option polling timers and retries expire and the trunk is seen as down, there'll be no delay at all.
