If the admission test of the destination succeeds, the second pass -
the reply pass - is initiated. During this pass, an
establish accept message is forwarded from the destination upstream.
As the message travels upstream, each node calculates the
feasible delay of packets on that node. By the feasible delay
on Node 
, we denote the maximum cumulative
delay a packet can experience on nodes upstream without missing the
end-to-end delay at the destination. In other words, a packet may miss
its deadline at one or more destinations if it is delayed for more
than before arriving at Node . The
feasible delay on is defined as follows:
where b denotes the number of successors of .
As the establish accept packets generated at the destinations
travel up the multicast tree, they carry with them the feasible delay
of the last node they traversed. Each node 
waits for the
messages from all its successors 
before determining the local feasible delay according
to Equation 3, and forwards the message with the new value to
the next node upstream. At each node , the successor with the
minimum value for its feasible delay determines the direction of the
critical path for the subtree with node as root. During the
reply pass, every node is informed about the direction of the critical
path by comparing the feasible delay values of incoming messages.
Each node informs the next node upstream about the length
of the critical path (in numbers of links) starting at
. In summary, upon receiving the establish message from all
successor nodes, the message from successor 
having the
entries ( 
), the node performs
the following four steps.
with the minimum value 
.
. ,
+ 1.
) to the next node upstream.
Let 
. Then by equation (3), it follows that
We note that the link from 
to 
is a critical path,
and is set as critical node, 
. Thus,
We note that the feasible delay 
at the source node indicates
how much additional delay can be introduced along the critical path of
the connection without missing end-to-end deadlines at the receivers.