Energy-Delay Optimal Routing Algorithm (EDOR)

Using shortest-path routing to choose a replica of a data object in the IP over WDM network can result in increasing the power consumption of the network as the shortest path may involve more hops (hence IP ports) and furthermore more router ports and transponders may be required to establish a new virtual link if enough capacity is not available on existing virtual links on the shortest path. Therefore we propose a new routing algorithm, Energy-Delay Optimal Routing (EDOR), to route traffic demands to data objects. EDOR aims to minimise the energy consumption

while maintaining QoS (propagation delay). The flow chart associated with the EDOR algorithm is shown in Fig 5-18.

Fig 5 - 18: EDOR algorithm flowchart

In this algorithm, all the traffic demands between data centres and nodes are reordered from the highest to the lowest and an empty virtual topology is created. A traffic demand is then retrieved from the ordered list. All the available paths to all the required data centres are checked. If more than one path has sufficient capacity, the required data centre with the shortest available path is selected in order to reduce the propagation delay. If sufficient capacity is not available in the virtual topology, a new virtual link is established between the node and the data centre with the minimum number of hops in order to minimise the power consumption by reducing the number of transponders in intermediate nodes. After routing the traffic demand, the remaining capacity on all the virtual links is updated. The above process is

repeated for all the traffic demands. After routing all the traffic demands on the virtual topology, the total power consumption of the network is calculated.

Under the EDOR algorithm the same data object exists in multiple data centres, so the data centre routing problem becomes a form of “anycasting” which introduces a degree of freedom in selecting the destination with the minimum energy consumption. The multi-hop bypass heuristic proposed in [58] is a unicasting algorithm, i.e. the destination is pre-determined and only the most energy efficient route is selected.

5.3.3 Simulation and Results

In this section we identify the impact of the replication scheme in the NSFNET network with 5 data centres. The data centre traffic demand and other parameters are similar to the assumptions in Section 5.2.2.

Fig 5 - 19: The power consumption of the IP over WDM network with optimal locations of data centres under the non-bypass heuristic with and without replication

In Fig 5-19, the performance of the replication scheme under shortest-path routing with the non-bypass scheme is evaluated. Simulation results are reported under the optimal data centre locations ((5, 6, 8, 10 and 13) obtained from running the model in Section 5.2.1 for the non-bypass heuristic) considering data centre traffic and regular traffic. The optimal selection of data centres is determined to replicate data objects through the use of the MILP model in Section 5.3.1.

Implementing the replication scheme has resulted in an average power saving of 28%. This significant saving is due to the reduction in the number of hops and the distance between data centres and nodes (as under replication objects are available in multiple data centres).

Fig 5 - 20: The power consumption of the IP over WDM network with optimal locations of data centres under the EDOR algorithm and the multi-hop bypass heuristic with shortest path

algorithm with and without replication

Fig 5 - 21: Reduction in the power consumption of the IP over WDM network with optimal locations of data centres under the EDOR algorithm and the multi-hop bypass heuristic with

shortest path routing

The results under EDOR with the optimal data centre locations (3, 5, 8, 10 and 12) have also been evaluated using the model in Section 5.2.1 for the multi-hop

bypass heuristic considering data centre traffic and regular traffic. The optimal selection of data centres to replicate data objects was determined from the MILP model in Section 5.3.1. Fig 5-20 gives the power consumption of the IP over WDM network with data centres with and without the replication scheme. It is clear that implementing the replication scheme has resulted in power savings under both shortest-path routing and the EDOR algorithm. The difference between the EDOR algorithm and the multi-hop bypass heuristic with shortest-path routing can be seen in Fig 5-21 where we show the power saving introduced by the replication scheme under the two algorithms compared to the multi-hop heuristic without replication.

While the EDOR algorithm achieves an average power saving of 4.5%, the multi-hop bypass heuristic with shortest-path routing average power saving is limited to 3.7 %. This is because the EDOR algorithm allows more traffic demands to share the capacity on common virtual links and therefore a smaller number of new virtual links need to be established. It also routes using a minimum hop criterion (minimum number of IP ports, switches, transponders and multiplexers and demultiplexers) when there is no sufficient capacity on established lightpaths. Between 04:00 and 08:00, the difference between the two algorithms reaches its peak as during this time period the average traffic demand is the lowest, and is lower than the capacity of a wavelength. The EDOR algorithm uses all available virtual links with sufficient capacity therefore more traffic demands share the capacity on common virtual links.

Fig 5 - 22: Average propagation delay of IP over WDM network with optimal locations of data centres under different algorithms

Fig 5-22 gives the average propagation delays under different algorithms. The multi-hop bypass heuristic without the replication scheme gives the upper bound on the propagation delay. The multi-hop bypass heuristic with the shortest path algorithm results in the lowest propagation delay (2.59 ms). With the EDOR algorithm, the propagation delay has not increased significantly (the increase is less than 0.2 ms, i.e. less than 8% compared to the lowest propagation delay) maintaining the QoS. It should be noted that while the propagation delay of the multi-hop bypass heuristic without replication and with the shortest-path algorithm are almost constant in a 24 hour period, the average propagation delay of the EDOR algorithm fluctuates slightly as the routing paths are dynamic.

5.4 Applying Renewable Energy in the IP over WDM