TL;DR
This paper introduces BA-VNE, a bandwidth-aware multi-domain virtual network embedding algorithm designed to meet the intense communication demands of the Internet of Vehicles (IoV). By combining a centralized network virtualization architecture with an optimized Particle Swarm Optimization (PSO) algorithm, BA-VNE successfully prioritizes high-bandwidth links while simultaneously reducing embedding costs and latency.
What Problem Does This Paper Address?
The rapid rise of the Internet of Vehicles (IoV) and autonomous driving requires vehicles to independently gather massive amounts of real-time environmental data. Tasks like predicting road conditions, calculating distances to surrounding cars, and perceiving traffic light changes demand instantaneous, high-bandwidth wireless communication.
As shown in Figure 1, this creates immense pressure on underlying network infrastructures.
Network Virtualization (NV) allows multiple virtual networks to run on a shared physical substrate, offering a flexible way to manage these resources. However, mapping these virtual network requests (VNRs) across multiple physical domains—known as Virtual Network Embedding (VNE)—is highly complex. Existing multi-domain VNE algorithms focus almost exclusively on minimizing financial costs or maximizing request acceptance rates, leaving bandwidth constraints secondary. This paper addresses the critical gap of prioritizing bandwidth allocation in multi-domain wireless environments.
Methods
The authors propose the Bandwidth Aware Multi-Domain Virtual Network Embedding (BA-VNE) algorithm. It utilizes a centralized architecture featuring a global controller and multiple local domain controllers.
Key components of the methodology include:
- Information Abstraction & Link Aggregation: To prevent high signaling overhead, local controllers do not upload their entire topology. Instead, they aggregate internal paths and only upload critical data (boundary nodes, available resources, and unit prices) to the global controller.
- Bandwidth-Filtering Objective Function: Before node selection, the algorithm filters out any substrate links with bandwidth lower than the domain's average link bandwidth. This ensures only resource-abundant paths are considered.
- Particle Swarm Optimization (PSO) with Mutation: The global controller constructs a global candidate network and applies a discrete PSO algorithm to find the optimal pre-mapping scheme. Crucially, the authors introduce variable mutation factors into the standard PSO to prevent the algorithm from getting trapped in local optima.
The overall execution flow is detailed in the algorithm flowchart:
Datasets
The algorithm was evaluated using synthetic network topologies generated via the standard GT-ITM tool. The simulation setup included:
- Substrate Network: 4 physical domains, each with 30 nodes and 2 boundary nodes. Substrate node CPU resources were uniformly distributed between 100 and 300 units, and link bandwidths ranged from 1000 to 3000 Mbps.
- Virtual Network Requests (VNRs): Arrival rates simulated a Poisson process, with CPU and bandwidth demands uniformly distributed between 1 and 10 units.
Key Insights and Ideas
- Pruning via Averages: Instead of treating bandwidth as a post-mapping metric, BA-VNE uses the domain's average bandwidth as an active filter. This drastically narrows the search space to high-performing infrastructure.
- The Load-Balancing Side Effect: By intentionally routing data through links with abundant bandwidth, the algorithm naturally achieves load balancing. This prevents network bottlenecks and unexpectedly improves the long-term request acceptance rate.
- Heuristics Beat Greed: Traditional greedy strategies (like mapping the cheapest links first) fail in complex multi-domain scenarios because they lack a global perspective. Introducing a modified meta-heuristic like PSO yields significantly better resource coordination.
Key Results
The BA-VNE algorithm was compared against several representative VNE algorithms (VNE-PSO, LID-VNE, MC-VNM, and MP-VNE).
- Superior Bandwidth Selection: BA-VNE consistently selected substrate links with average bandwidths significantly higher than the baseline domain averages and competing algorithms (see Figure 6).
- Minimized Cost and Latency: BA-VNE achieved the lowest comprehensive embedding cost across varying numbers of virtual nodes. It also maintained the lowest average mapping delay (under 500 ms), whereas greedy approaches exceeded 850 ms.
- Higher Request Acceptance: As the network load scaled up, competing algorithms saw their VNR acceptance rates plummet to around 30%. Thanks to its load-balancing properties, BA-VNE stabilized at a robust 60%.
Limitations and Future Work
While BA-VNE performs exceptionally well in simulations, the paper relies on a centralized global controller architecture, which introduces a potential single point of failure and scalability concerns for massive global deployments. Furthermore, the simulation model assumes static substrate nodes and does not fully capture the high mobility and rapid topology changes characteristic of real-world vehicular networks.
The authors note that future iterations could expand this bandwidth-aware framework to support other high-throughput services, such as large-scale video data backups.
Summary updated