Replica Dissemination and Retrieval Protocol Overhead

To quantitatively compare the SID message overhead with the other presented RR solutions, we have measured the overall number of messages required to distribute and find replicas in the reference scenario. Figure 1 reports the experimental results obtained while increasing the number of searches. Each point represents the average of 20 simulations. SID parameters i and s are set to 12. This value has been experimentally chosen by considering that, when i and s are greater than the diameter of the considered dense MANET (12 hops), the accuracy overcomes 85% (accuracy results). k-DID exploits the same number (12) of disseminated IRPs, so to realize an actual deployment scenario where it makes sense to perform a comparison between message overheads.

Figure 1: Message overhead for the four presented RR solutions.

Figure 1 shows how QF produces a rapidly growing amount of message exchanges also for a limited number of searches. As expected, k-DID imposes a high overhead for IRP placement (about 4 messages per node in the dense MANET), while its search phase demonstrates to be very effective. SID exhibits linear growth in overall message overhead, by imposing a lower number of IRP distribution messages than QF and k-DID and only a slightly greater number of RR messages than k-DID. Both k-DID and SID require low memory occupation on IRP-hosting nodes, since they diffuse IRPs only on a limited node subset (see the concise comparison among RR solutions in Table 1). IF represents a lower bound in terms of communication overhead, but it is a practically unviable solution because it requires all nodes in the dense MANET to store IRPs about all replicas of all available resources.