5.5 Communication Network Models
5.5.2 Network Scenarios
Figure 5.6: The main communication network model spread out across the area of New England. Every substation consists of several Relay Agent nodes, network switches, and routers. They are interconnected via wide area networks that run along the transmission lines.
5.5. COMMUNICATION NETWORK MODELS
Figure 5.7: The IEEE 39-bus system with the communication network infrastructure represented as on overlay. Every ellipse denotes a local area network at a substation, which connects local Relay Agents.
Dedicated Infrastructure - No Background Traffic (noBT)
It was assumed that the communication infrastructure was dedicated to the Relay Agent communication and not utilised to transfer any other data traffic. While this scenario is less likely mainly because of the costs associated with a dedicated infrastructure, the obtained results for this scenario indicate the best possible performance that can be achieved without changing the communication infrastructure or technologies. This scenario was named ”No Background Traffic” (noBT) as there is no other traffic present than the one created by agent communication.
Competing Background Traffic (BT)
A significantly more realistic scenario is the use of a converged communication network. In this scenario the network is used to carry data that originates from all kind of different sources such as office applications, substation monitoring equipment, surveillance videos, and IP telephony. Therefore, the Relay Agent communication has to compete with additional traffic, which is called background traffic (BT). An average link utilisation of around 90% was assumed, which meant that the Relay Agents had to compete with a 1Mbit/s background traffic. For some simulations this link utilisation was also increased further to investigate its impact.
The background traffic was modeled in OPNET as analytical flows (ip traffic flow model shipped with OPNET) between all inter-substation links. These flows impact the performance of the explicitly modeled agent traffic by introducing additional queuing delays to the network devices.
Quality of Service (QoS)
A Quality of Service (QoS) strategy was implemented to help the agent communication compete against the background traffic introduced in the previous scenario. In this context QoS refers to the ability to give different priorities to different types of data traffic. Because the proposed supervision scheme is time-critical, it is desirable to treat its traffic more favourable over all other traffic.
The implemented QoS is based on the Differentiated Services (DiffServ) architecture with class-based weighted fair queueing (CBWFQ) and a low latency queue (LLQ) [113, 114]). In a DiffServ architecture network traffic is classified and treated differently by network equipment such as routers and switches. Figure 5.8 shows an example of a
5.5. COMMUNICATION NETWORK MODELS
QoS-enabled networking device with 3 queues. The received data packets are put in one of the FIFO queues according to their classification. The scheduler decides from which queue the next packet is put on the medium. The packets may be classified by different parameters, such as traffic classes, input interfaces, source addresses, and destination addresses. Traffic classes might honor fields in protocol headers that represent DiffServ markings. For example, the IPv4 header contains a 6-bit DiffServ Code Point (DSCP) value that may be used for traffic classification.
Figure 5.8: Example of a QoS-enabled networking device with 3 queues. Data packets are put into different FIFO queues based on their classification. The scheduler decides from which queue the next packet will be put on the medium.
The routers at the substations honored the DSCP values to classify the traffic into 8 type of service classes (see Table 5.2), which are defined as a QoS profile in OPNET. Each traffic class has its separate FIFO queue and the scheduler uses OPNET’s weighted fair queueing (WFQ) scheduling technique to decide from which queue the next packet will be forwarded. This technique ensures that every queue has an average data forwarding rate that is proportional to its assigned weight. In addition to WFQ, the interactive voice class was defined as a low latency queue (LLQ). A LLQ introduces a strict priority queue into WFQ and traffic in this queue gets the highest priority. Only if the LLQ is empty, are other queues allowed to be emptied according to the WFQ mechanism. The total buffer size for each router was set to 1 Mbytes.
The traffic of the agent communication was given the DSCP value of 48 (i.e. interactive voice class) and all other background traffic was marked as the background class with a DSCP value of 8. There was no need to separate the background traffic into different service classes because only the agent communication was marked for the LLQ, which is always given the highest priority. However, the unused service classes could be used if additional agent applications were to be introduced in the future.
Table 5.2: Implemented traffic classes and their DSCP values and weights.
Class name DSCP value Weight
Best Effort 0 1 Background 8 10 Standard 16 20 Excellent Effort 24 30 Streaming Multimedia 32 40 Interactive Multimedia 40 50 Interactive Voice 48 LLQ Reserved 56 70 Scenario Variations
While all 3 scenarios (noBT, BT, QoS) were simulated with some common scenario vari- ations, other variations were only simulated for the most practically relevant scenarios in order to reduce the overall simulation time to obtain the results. For example, all 3 scenarios were simulated with UDP and TCP as transport protocols, message sizes of 250 and 350 bytes, peer-to-peer and client/server communication approach. On the other hand, communication link outages, additional communication links, a wider range of message sizes, reduced peer selection, and DMA at different substations were mostly simulated for the QoS scenario and peer-to-peer strategy.