What is it?

“Spooky action at a distance,” a famous phrase by the physicist Albert Einstein, describing the effects of Quantum Entanglement. Local action on a particle determines the state of a far distant entangled particle.

We devise autonomic agents that use local information (internal and direct neighbors) to enforce global topological properties of large-scale Wireless Mesh Networks (WMN). This design choice, which we denominated Smart agents, makes our WMNs highly resilient to failures, turning them an excellent fit for unplanned, large-scale WMN deployments. Moreover, the network formation does not rely on manual input of configurations or a pre-determined and centralized controlling mechanism.

Both capacity limitations at large-scale and network control inflexibility due to the use of complex distributed routing protocols are critical limitations that hinder the application of WMNs at large scales. Moreover, SDN-based WMNs have not been practical for large-scale deployments.

RESEARCH

In this research project, we apply the principles of Autonomic Computing to direct the formation of wireless mesh networks with particular properties that will:

1. Support the general applicability of centralized network control mechanisms (SDN) to large-scale WMNs, and
2. Foster network throughput capacity increase by allowing flow parallelism advent from frequency diversity enforcement and advanced SDN-based traffic forwarding.

The autonomic adaptations at the physical and link layers dramatically reduce the effort on network creation and operation, turning our model of large-scale WMN implementation tractable to unplanned and unbounded settings such as disaster network scenarios and community wireless networks.

The outcome of this autonomic formation process is a set of partitions with controlled density (an upper bound on the degree of all nodes), and controlled diameter. The latter is a critical factor to bound the latency between nodes of a partition, allowing the practical application of per partition SDN controllers. The former maintains capacity given the density bounding, even in cases in which an unplanned deployment turns the total node density too high. We also rely on a form of network-based leader election on partitions that allow the convergence of the formation process to a stable topology outcome. We currently devised two types of autonomic agents to organize the nodes into partitions (autonomic self-organization) and to interconnect the partitions (self-healing). Both agents are also self-configuring and self-stabilizing (converge to stable settings).

In the following research developments, we aim at turning partitions into a form of meta-agent, which is self-optimizing. The self-optimizing partitions can choose frequencies that minimize overlap with nearby partitions, and adapt partition parameters (node degree, partition diameter) to adapt to changing conditions.