Voice Communication

We add up the number of voice packets which are transmitted through the ad hoc network as Figure 5.5 illustrated. We can find out that GPMS, ACT, and CLAPS all have the similar performance, thus P2P solution can be demonstrated in an efficient delivery with scalability.

Especially, GPMS has the performance approximated to an ideal EN(n) as in Equation (9), because every peer only copies once for its rear peer and every intermediate node only copies once for the destination.

The packet failure rate is a major metric of live streaming, and it is defined as the number of multimedia chunks that lose or arrive before playback deadlines over the total number of multimedia chunks. As Figure 5.6 illustrated, GPMS has the lowest packet failure rate among all compared models. However, the large network leads to the long routing path, which leads to high packet failure rate. ACT and CLAPS have the high packet failure rate, because the collision happens in the overlap of spanning trees when many peers simultaneously speak.

Especially CLAPS overlay tends to share the overlapping routing path, thus its packet failure

100 120 140 160 180 200

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Churn rate

Completion time (seconds) .

ORION MADPastry GPMS

Figure 5.4: Download completion time vs. the churn rate.

rate raises seriously with scalability.

The simulation result about mobility is curved in Figure 5.7, and we can discover that GPMS has the lowest packet failure rate among the compared models. The cross-layer scheme detects peer movement to avoid the far routing path. The cross-layer scheme always performs time sensitivity of neighborhood to keep proximity.

As Figure 5.8 illustrated, the proposed GPMS has the lowest packet failure rate among the compared schemes. Because the cross-layer design speeds up recovery latency and improves streaming stability, voice packets can be redirected through multiple intermediate nodes. FPRT is always updated to recover backup routing path, but ad hoc routing disconnection or rediscovery on OLSR may be conflicted with overlay recovery in ACT and CLAPS. When the churn rate is higher than 0.8, three models work impossibly, so the broken delivery path hurts VoIP service seriously.

The amount of voice copies .

ACT CLAPS GPMS Ideal EN(n)

Figure 5.5: The amount of voice copies increases with the number of peers.

0.0

Figure 5.6: Packet failure rate vs. the number of nodes.

As Figure 5.9 illustrated, in GPMS, even if the overlay changes, the moving peer always influences only neighboring peers. Therefore, the low speed does not cause the high packet loss.

However, the high speed changes the overlay drastically, which leads to the continuous and frequent packet loss. Although packet failure rate is rose with moving speed, GPMS still keeps the sequential receipt. Because the hop-by-hop routing path inherits the unidirectional stream, the voice arrives in order even if multiple peers speak simultaneously.

5.3. Video streaming

In order to build a simulated benchmark, we characterize the delivery principle, the overlay construction, and the streaming factors on Joost [36] referred to MP2PS. The video packet size is 1024 bytes, and the data rate of streaming is 450 kbps (constant bit rate). As Figure 5.10 illustrated, the network stacks of compared schemes are described to differentiate between

0.0

Figure 5.7: Packet failure rate vs. the moving speed.

0.0

Figure 5.8: Packet failure rate vs. the churn rate.

network features.

As Figure 5.11 illustrated, the proposed GPMS has the lowest packet failure rate among the compared schemes. When the number of nodes increases to 120, the failure rate is 0.07, so the proposed GPMS is demonstrated with scalability. However, MP2PS cannot support high scalability because the overlay forwarding is not mapped into the ad hoc routing. The larger number of peers is relative to the more routing paths, so the flooding query degrades the streaming continuity in Smart Gnutella.

In wireless network, the mobility problem leads to a difficulty of keeping a continuous playback. As Figure 5.12 illustrated, the cross-layer design can be appropriate for the mobility.

Due to the overlay maintenance and overlay proximity of GPMS, it can maintain a low failure

0

Figure 5.10: The compared network stacks.

rate and low overhead when mobile peers move. Smart Gnutella also has stable performance against the moving speed, because the moving speed affects the flooding scheme slightly.

However, DHT originally is designed for wired network, so MP2PS is unable to detect peer movement, such that the peer cannot find its nearest neighbor to forward stream. Under the high moving speed, GPMS highlights the advantage for mobility.

For live streaming, the peer churn leads to a difficulty of keeping a stable overlay. As Figure 5.13 illustrated, the proposed GPMS is suitable for a dynamic MANET, because the cross-layer design speeds up recovery latency and improves streaming stability. FPRT is always updated to recover backup routing path, but ad hoc routing disconnection or rediscovery on AODV may be conflicted with overlay recovery in Smart Gnutella and MP2PS. Therefore, the packet failure rate is only 0.25 in GPMS when the churn rate is 0.5, however, that is 0.44 and 0.78 in Smart Gnutella and MP2PS respectively. GPMS is suitable for dynamic network

0.0

Figure 5.11: Packet failure rate vs. the number of peers.

0.0

Figure 5.12: Packet failure rate vs. the moving speed.

with high churn rate.