Figure 13. Throughput under different packet injection rates of the 8×8 Mesh-of-Torus network and mesh network in different flow patterns.
 
Serial number
Time interval
Injection rate
(packets/node/s)
1
0.000001
1,000,000
2
0.0000009
1,111,111
3
0.0000008
1,250,000
4
0.0000007
1,428,571
5
0.0000006
1,666,666
6
0.0000005
2,000,000
7
0.0000004
2,500,000
8
0.0000003
3,333,333
9
0.0000002
5,000,000
10
0.0000001
10,000,000
Table 6. Control table between simulation serial and packet injection rates
 
6.3 Comparison between Mesh-of-Torus network and mesh network
Latency: Figure 12 shows the statistic end-to-end latency simulation results for the 8×8 Mesh-of-Torus network and mesh network with a 256 bytes packets size and different flow pattern [28]. From the Figure 12 we can see that the saturation point of the Mesh-of-Torus network is higher than that of mesh network, that is because that long links that used to reduce the network diameter can enable packets to reach the destination node in a higher speed. Besides, the latency of the Mesh-of-Torus network is naturally lower than that of mesh network because that when the diameter of the network is reduced, the latency of packet transmission will decrease simultaneously.
It will also help to support the fact that both mesh network and Mesh-of-Torus network are in line with real world data center.[O4] 
Throughput: Figure 13 have shown statistic of throughput of in different packet infusion rates and the simulation results for the 8*8 Mesh-of-Torus network and mesh network with a 256 bytes packets size in differentiated flow pattern. The network will basically attain the same throughput when the network is not saturated. However, when the network is saturated, the throughput of Mesh-of-Torus is obviously higher than that of esh network. More detailed infusion rates for the test have been listed in table 6.
Figure 13. Abscissa represents infusion rates for different groups, and the coordinate represents for the throughout for the entire network. And the unit represents for the packets number.
Link Load Distribution: Figure 14 have shown the statistics of the link load distribution in the same network model deploying Mesh-of-Torus network and mesh network [30]. Long links in Mesh-of-Torus network can uniform the load distribution better by sharing some of the traffic inside different parts in the network. And simulation results also prove the excellent performance of Mesh-of-Torus network in traffic distribution balancing as well as latency reduction. Based on all we have discussed above, Mesh-of-Torus network is much more suitable for the data center whose local traffic is much heavier than global traffic, so the higher practicality in real world data center employment since data centers in real life are common in the use of local traffics.