Network digital twins offer an efficient, cost-effective way to assess how a network will behave under different operating conditions. For example, consider the enterprise network for a bank which consists of manned teller stations, ATMs, web-portal for customer access on the frontend and a network of primary and backup servers at the backend which process a large volume of transactions 24 hours a day. These networks are prime targets for increasingly sophisticated cyber-attacks. Given the mission-critical nature of these networks, it is of paramount importance that the networks be cyber resilient, i.e., the network should be able to thwart cyber-attacks, protect the confidentiality of data, and continue to provide service even when it is under attack. Launching cyber-attacks against a live network to evaluate its cyber resilience is clearly not an option. Network digital twins or network simulations provide a convenient, cost-effective method to evaluate cyber resilience in a safe, controlled environment without the danger of causing harm to the network itself, under a large number of operational conditions. Similarly, other network performance criteria besides cyber resilience can also be evaluated using network simulation.
But simulation of networks, with a large number of mobile and stationary devices communicating with each other to transmit all the control and data traffic can be very computation intensive. As shown in Figure 1, the time it takes to execute a simulation increases dramatically with the number of devices in the network. Additionally, to generate data for meaningful “what-if” analysis, a large number of simulation runs are required. Hence, to be of practical use to planners and sysadmins, the digital twins must be able to produce actionable reports rapidly – this directly requires that the underlying network simulation models can be used to rapidly evaluate a number of alternatives and assess the most suitable network setup to meet given requirements.
The need for fast network simulations arises from another important application: If a simulator can run in hard real-time, i.e., the simulation takes 1 unit of execution time to simulate the operation of a corresponding physical network for the same duration, then the digital twin can interface with real applications and live hardware in a Live, Virtual, Constructive (LVC) environment. This provides a means for testing real applications on a realistic testbed (which exhibits the same end-to-end delays and behaviors as the physical network) and for incorporating live hardware in an LVC environment where live and simulated devices interact with each other (thereby enabling network evaluation at dramatically reduced costs because a smaller number of real devices than in a purely physical network are needed in an LVC environment). As Figure 2 shows, SCALABLE’s network simulators, EXata and QualNet, can run in hard real time for large WiFi networks with as many as 1,500 devices under realistic operating conditions.
Impact of Slow Execution Speed
To get around the problem of simulations taking a long time, legacy simulators have generally relied on using low fidelity models of protocols and network components, using models of smaller sizes and extrapolating simulation results to larger networks, or decomposing the network operation into different phases and analyzing each phase independently. These methods can reduce the simulation time, but this increased speed comes at the cost of accuracy of results. Simulation models which use one or more of these approaches do not capture the network operations in sufficient detail and the simulation results cannot be applied to real networks with a reasonable measure of confidence.
High-fidelity, “at-scale” simulation models are essential for accurate simulation results. In addition to being of practical use to planners, the simulator should be capable of running simulations at a fast speed. Hence, fidelity, scalability and speed are the most important attributes of a simulator for it to be useful in assessing the performance of networks.
How SCALABLE Achieves Speed, Fidelity and Scalability
SCALABLE’s network simulators, EXata and QualNet, use state-of-the art techniques to run simulations of large networks at a fast speed without compromising accuracy. EXata and QualNet are based on a highly specialized kernel and efficient memory management techniques which result in very high simulation speeds. Furthermore, our simulators exploit the computing power available on multi-processor platforms by employing efficient parallel discrete event simulation techniques and smart partitioning, i.e., dividing the workload optimally among the processors. As a result, EXata and QualNet can run on laptops and desktop computers to simulate large networks fast, without compromising the accuracy of simulation results. SCALABLE’s technology creates a network digital twin, representing the entire system network, the various protocol layers, application layer, physical layer, and devices. EXata also includes a low-skew synchronization kernel to connect with live applications, which run on the network digital twin just as they would run on real networks.
As Figure 3 shows, EXata and QualNet can simulate on-the-move (wireless) networks with up to 1,500 nodes faster than real-time on a standard server class hardware with 8 cores. This allows several simulation runs of medium-to-large networks to be completed in a matter of hours. The figure also shows the increase in speed with the number of cores.
Even for much larger networks, simulation execution on these servers is very fast and several simulation runs can be completed over-night. For example, as Figure 4 shows, simulation of a 10,000 nodes on-the-move network, with a simulation horizon of 300 seconds, completes in about 1,800 seconds on a machine with fewer than 16 cores! For the same simulation horizon, a 5,000 nodes network simulation takes about 670 seconds, which is only about 2.25 times the total simulation time. Even on a desktop with 8 cores, the 5,000 nodes network simulation takes about 830 seconds, which is less than 3 times the total simulation time.
In summary, network digital twins created using SCALABLE’s technology can be used to simulate very large networks under realistic operational conditions to produce actionable results on a continuous basis. Analysts no longer need to compromise on model fidelity (and hence accuracy of the generated results), but now have access to network simulation tools that can provide model accuracy and scalability without compromising on execution speed!
 For details of EXata’s and QualNet’s performance, refer to SCALABLE’s white paper, Fast and Accurate Simulation of Scalable Networks