The Complexity of the Edge-Computing Infrastructure

The rise of local processing capacity at the edge is driven by numerous advantages critical for future processing tasks. These benefits include enhanced security and reliability, as well as reduced latency and energy consumption. Managing the edge infrastructure, known as the edge continuum, introduces a dynamic computing landscape. Within this continuum, we consider a network of Edge Micro Data Centers (EMDCs). Intelligence is distributed across these edges, creating a decentralized environment. This decentralization allows the edge to be more autonomous and precise in local decision-making within a regional context, reducing dependency on a central coordination point. This is particularly crucial for real-time applications such as autonomous driving or the monitoring and control of smart grids. The edge infrastructure faces greater challenges in stability and performance due to stricter latency and autonomy requirements, distribution across multiple sites, its locally limited size, multi-tenancy, and multiple operators, along with local management of concurrent and asynchronous components.

ACES as agent-based modeled system.

Central to our conceptual approach is the use of swarm agents to represent demand and supply entities. Demand swarm agents simulate workload behaviors at the pod level, ensuring optimization at that level. Conversely, supply swarm agents model node dynamics. These agents collaborate within an EMDC environment, orchestrating processes such as workload placement, storage management, and caching optimization. Using these agents, we can employ exemplary swarm algorithms, e.g., the hormone and ant algorithms, to achieve the desired system functionality. For instance, demand swarm agents use synthetic hormones to communicate their requirements and priorities. Supply swarm agents detect these hormones to make informed allocation decisions. The ant algorithm dynamically optimizes workload-to-node assignments by mimicking the foraging behavior of ants, depositing pheromones to guide subsequent decisions.

Demand and Supply Swarm Agents.

The interaction between demand and supply swarm agents is orchestrated through swarm intelligence algorithms. Demand swarm agents autonomously identify the most suitable nodes for workload placement, while supply swarm agents determine the optimal workloads to process based on available resources and capacity. This collaborative decision-making process allows the system to efficiently allocate workloads to nodes, optimizing processing, latency, and resource utilization.

Our agent-based model is designed to exhibit autopoietic characteristics, promoting self-organization, regeneration, and regulation within the edge continuum. As demand and supply agents interact and adapt to changing workloads and resource availability, the system as a whole displays emergent behaviors that enhance its resilience and efficiency.

Demand Swarm Agents:

An application is split into a set of services 𝒮 that are represented as a set of related pods P^s={p^s₁,p^s₂,… } with sS as the specific service. Each service sS is defined by a compilation of resources Rs which prescribes the processing steps necessary to compute the individual pods. The pod pjspjs​ can choose which of the suitable nodes Nⁿ_i to use for each necessary process step P^r.

Supply Swarm Agents:The EMDC E contains several sets of nodes or nodes, consisting of different types of resources Nr= {N¹_r, N²_r, …}, where r is/are the resource type(s). A node with different resources presents a typical EMDC node, whereas a node with a single resource presents, e.g., a CPU that is part of a pool of resources. In the course of this work, we consider the following resources along with their respective capacities: CPU, FPGA, RAM, and NVMe. Each resource N^r_i has a queue N^r_i.