We assessed the quality of the three-pass protocol described in this work we performed a suite of simulation experiments. The goal of these experiments was to determine the effect of this protocol on the Call-Blocking Probability, that is, the probability of a request for connection establishment being rejected.
In the experiment suite, we simulated the establishment and teardown of one-to-many uncompressed CD-audio connections with a varying numbers of destinations. Specifically, the traffic specification for the connections is that of constant-rate streams with maximum packet size of 1KB and minimum interarrival time of 13ms, for a local bandwidth requirement of 0.7Mb/s. The end-to-end delay requirement is 188ms.
The underlying network consists of nodes connected by OC-12 level links (620Mb/s). The nodes use RCSP schedulers to control the traffic on the outgoing links. The schedulers have 6 priority queues, which are configured so as to have delay bounds 5, 14, 27, 40, 66, and 100 packet transmission times, respectively. According to the acceptance test criteria described in Section 2.3, each outgoing link can carry up to 4 connections at priority 1, up to 6 connections at priority 2 and above, 8 connections at priority 3 and above, 9 connections at priority 4 and above, 10 connections at priority 5 and above, and 12 connections at priority 6 and above. Each node has a 30MB buffer to store the packets temporarily in the traffic regulator and the priority queues.
Requests for connection establishment form a Poisson process with rate
, and life times for connections identically distributed with
the common distribution being exponential with mean 
. In the
following we will keep fixed to one normalized time unit, and
will let vary.
An establishment request for a connection can be rejected for three reasons; first, the end-to-end delay violates the guarantees at a destination, second, a node on the route cannot allocate enough buffer space, and third, the bandwidth on a link is exceeded. If one of the destinations or a node on its path rejects the connection, all temporarily reserved are freed.
We note at this point that if a connection is rejected for lack of bandwidth, this is not caused by inadequate allocation of resources, but is issues of inappropriate routing. Since we are no addressing routing in this work, in the following experiments we ignore establishment requests that are rejected because of lack of available bandwidth. The performance metric in these experiments is the average number of accepted connections in the system at any time in steady state. The experiments were performed for two topologies, depicted in Figure 3. Individual connections were guaranteed by randomly locating senders and receivers, and connecting them with shortest routes from the senders to the receivers.
Figure 3: Network Being Simulated.