In order to optimize the timeslot size setting in the SynOBS network, we initially consider the minimized nomalized offered load (bursts/timeslot) given to the entire network (the timeslot size setting which results in the minimum nomalized offered load).
Extend the normalized offered load (ϕ) generated by a data source given in equation
5.17, the normalized offered load given to the entire network can be calculated by
ϕnetwork =
S
X
i=1
ϕi (6.1)
where S is the total number of data sources in the entire network.
Figure6.1 shows a network configuration that is used to study the effect of timeslot size in the SynOBS network. As shown in the Figure, the network consists of five core nodes (labeled as 1 to 5), four groups of data sources (labeled as A to D); each of which consists of ten data sources, and four groups of data sinks (labeled as E to H). Group A data sources
2 1 3 4 5 10 edge nodes A B C D E F G H core node
Figure 6.1: Experimental network
send their data to group E data sinks via node 2 and 4. Group B data sources send their data to group F via node 1, 2, and 4. Group C send data to group G via node 1, 3, and 4. And finally, group D data sources send data to group H via node 3 and 5.
Figure 6.2 shows the comparison results of (a) the calculated normalized offered load given to the network, and (b) the simulated burst blocking probability based on given target timeslot size setting. The results are from the network shown in Figure 6.1. Each of the data sources in the network has a packet arrival rate of 0.075 packets/µsec. The number of available data wavelengths is five. There are no FDLs available as an optical buffer in each core node. When a data packet is generated, it is then assembled to form a data burst according the algorithm as discussed in Section5.1. When the burst is assembled, the created data burst is sent out to its destination via core nodes as discussed earlier.
According to the graph, at a small target timeslot size setting, the normalized network offered load decreases as the target timeslot size increases. As the target timeslot size keeps increasing, the decreasing rate of the normalized offered load decreases. Until it reaches
0.001 0.01 0.1
10 20 30 40 50 60 70 80 90 100
Target timeslot size (usec)
Burst blocking probability
2 3 4 5
10 20 30 40 50 60 70 80 90 100
Target timeslot size (usec)
Network offered load (bursts/timeslot)
(a)
(b)
Figure 6.2: The comparison of (a) the calculated normalized offered load, and (b) the simu- lated burst blocking probability in the experimental network with balance offered load
the point where the minimum normalized offered load is obtained, then the normalized offered load increases as the target timeslot size increases. As the normalized network offered load decreases, the simulation’s blocking probability also decreases, because of lower traffic through the network. Moreover, the increase in normalized network offered load results in a higher blocking probability as well. Both the normalized network offered load and the blocking probability reach their minimum points around the same target timeslot size. In this case, target timeslot size is optimized to minimize the normalized network offered load, and we can effectively predict the target timeslot size that gives the lowest burst blocking probability.
0.001 0.01 0.1
10 20 30 40 50 60 70 80 90 100
Target timeslot size (usec)
Burst blocking probability
2 3 4 5
10 20 30 40 50 60 70 80 90 100
Target timeslot size (usec)
Network offered load (bursts/timeslot)
(a)
(b)
Figure 6.3: The comparison of (a) the calculated normalized offered load, and (b) the simu- lated burst blocking probability in the experimental network with unbalance offered load
Figure 6.3 shows the comparison results of the (a) calculated normalized offered load given to the network, and the (b) simulated burst blocking probability based on a given target timeslot size setting from the same network shown in Figure 6.1. This time, each of the data sources in group A and group B has a packet arrival rate of 0.05 packets/µsec, while the data sources in group C and group D has a packet arrival rate of 0.10 packets/µsec, thus creating an unbalance offered load to the studied network, where core node 2 and core node 4 receive less offered load than core node 3 and core node 5.
Again, according to the graph, at a small target timeslot size setting, the normalized network offered load and burst blocking probability decrease as the target timeslot size increases. As the target timeslot size keeps increasing, the decreasing rate of the normalized offered load and burst blocking probability decreases. Until they reach their minimum points, then the normalized offered load and burst blocking probability increase as the target timeslot size increases. However, in this case, the target timeslot size where the minimum network normalized offered load is obtained is smaller than the one where the lowest burst blocking probability is obtained. This is because the unbalanced offered load was given in the different network core nodes. In this case, nodes 3 and 5 (which receive traffic from data sources with higher packet arrival rate) receive more offered load than nodes 2 and 4 which results in more burst blocking in node 3 than node 2. Therefore, to reduce the burst blocking probability in the network, the timeslot size setting has to be adjusted to the given offered load in the more congested node as represented by the group C and group D data sources (0.10 packets/µsec). Consequently, the timeslot size setting that results in the lowest burst blocking probability is higher than the timeslot slot size setting where the the lowest network normalized offered load is obtained.
The discussion above shows that the timeslot size setting in which the minimum network normalized offered load is obtained may not result in timeslot size setting where the lowest burst blocking probability is achieved. Thus, optimizing timeslot size setting by minimizing the network normalized offered load can not be effectively used as the tool to determine the optimized target timeslot size setting that results in lowest burst blocking probability in the network.