LAKE – Swarm-based orchestration algorithms
| ID: R8 | Licence: Apache 2.0 |
Owner: LAKE |
Contacts: schranz@lakeside-labs.com |
| Short Description | A library of swarm-based orchestration algorithms, implemented in a Python-based simulation framework (COSMOS) for evaluating self-organizing scheduling algorithms in edge-fog-cloud continuum systems. Based on these principles, an Ant Colony Optimization algorithm runs as one of the Workload Managers of the MIRTO cognitive engine, allocating service chains on the nodes of the MYRTUS infrastructure without a central controller. | |
| Algorithms in simulation (COSMOS) | Workload Manager in MIRTO | |
|---|---|---|
| Key features | – Modular swarm coordination dynamics – Constraint-aware scheduling (ASP + metaheuristics) – Network-aware orchestration evaluation – Extensible API for custom algorithms – Mesa library integration for agent-based modelling | – One manager on every node of a Kubernetes cluster – Ant Colony Optimization: nodes with more successful allocations are preferred – Redirection of service chains between edge, fog and cloud clusters – Same interface as the other Workload Manager in MIRTOs |
| Require | – Python programming knowledge – Understanding of swarm intelligence principles – Mesa library dependencies | – Kubernetes cluster – MIRTO Agent and Deployer – Distributed Knowledge Base and Resource Registry |
| Provide | – Flexible experiment setup – Comparative analysis of scheduling paradigms – Real-time optimization capabilities | Allocation of the software components of a service chain to the nodes of the infrastructure. |
| Input | – Resource/demand agent configurations – Network topology definitions – Scheduling policy parameters | Service chains, with their software components and CPU and RAM requirements. |
| Output | – Performance metrics (latency, resource utilization) – Emergent behavior patterns – Optimization trajectories | Allocation map (software component to node), stored in the Distributed Knowledge Base. |
| User | Researchers and PoC software developers in distributed systems, swarm intelligence and edge computing. | MIRTO Agent and Deployer. |
| Benefits for the user | – Enables testing of decentralized algorithms in complex environments – Reduces prototyping costs for real-world deployments – Facilitates comparison of nature-inspired vs classical approaches | Decentralized, scalable and robust allocation of workloads, without a central controller. |
| Download | github.com/Incomprehensible/COSMOS | Please contact the developers listed above. |
| Role in MYRTUS | Comparative analysis of swarm algorithms with classical approaches. | Swarm-based Workload Manager of the MIRTO cognitive engine. |
| TRL@M0 | 1 | |
| TRL@M36 | 3-4 | |
General description
Algorithms in simulation (COSMOS)
Swarm intelligence, inspired by the collective behaviour of animals such as ant colonies, schools of fish and flocks of birds, relies on decentralized agents that follow local rules, without a central controller. The library models the edge-fog-cloud continuum with two kinds of agents: resource agents, representing the compute nodes with their available resources, communication costs, location, current load and energy consumption, and demand agents, representing the services (for example Kubernetes pods) that look for the most suitable resource. The algorithms are implemented in COSMOS, a simulation framework built on the Mesa agent-based modelling library.
Workload Manager in MIRTO
The swarm-based Workload Manager brings these principles into MIRTO. A manager runs on every node of a Kubernetes cluster and decides which software components of an incoming service chain can run on its node; the components that do not fit are passed to another node. The next node is chosen with an Ant Colony Optimization rule: each node has a pheromone value that grows with successful allocations and decreases when its resources are not enough, so that successful nodes are preferred without excluding the others, and overloaded nodes quickly lose priority. Pheromone values and allocations are stored in the Distributed Knowledge Base, and up-to-date resource information comes from the Resource Registry.
MYRTUS Extension/Contribution
Algorithms in simulation (COSMOS)
Within MYRTUS, the library provides a flexible environment to set up experiments on self-organizing scheduling in the continuum, to evaluate network-aware orchestration and to compare nature-inspired algorithms with classical approaches. Its API can be extended with custom algorithms, and local rules can also be designed with evolutionary algorithms using FREVO.
Workload Manager in MIRTO
The swarm-based Workload Manager has been implemented as a Python service with the same REST interface as the other Workload Manager in MIRTOs, so that the MIRTO Agent and the Deployer can use any of them. It has been tested with the MYRTUS infrastructure: simulations comparing different redirection strategies and orderings of the service chains showed energy savings with respect to random allocation, and functional tests confirmed the expected allocations. It has then been extended to several clusters: each cluster has its own pheromone value and a layer label, and a service chain that does not fit in an edge cluster is redirected to another edge cluster or to a fog cluster, from fog to fog or cloud, and from cloud only to another cloud cluster.
Technology Benefits and Innovations
Decisions are taken locally by agents that only communicate with their neighbourhood, so the approach scales as the number of devices and workloads grows, adapts quickly to changes in resource availability, network conditions and workload demands, and keeps working when single agents fail or become unavailable. Lightweight allocation and reduced dependence on fog and cloud resources also contribute to lower energy consumption, in line with the constraints of edge devices. Unlike centralized schedulers, there is no single point of failure and no need for complete global information.
Users
The simulation library is intended for researchers and proof-of-concept software developers working on distributed systems, swarm intelligence and edge computing, who need to test decentralized algorithms before real deployments. The swarm-based Workload Manager is suited to operators and integrators of large, decentralized continuum systems, where many devices host relatively short-lived services and scalability, energy efficiency and fault tolerance are the main priorities.
Application Domains
The approach fits decentralized systems that support self-organization, such as ad-hoc and mobile networks and large-scale edge infrastructures with hundreds of nodes. It is less suited to critical applications that require fully predictable behaviour, since the allocation emerges from the interaction of many agents.
References
- A. Stepanov, M. Schranz, A. Chepizhko, “Using Ant Colony Optimization for Scheduling Service Chains in MYRTUS Platform based on Kubernetes”, in Proc. PAAMS 2026.
- A. Ghasemi, M. Schranz, “Bottom-up resource orchestration in edge computing: An agent-based modeling approach”, in Proc. IEEE International Conference on Intelligent Systems, 2024.
- COSMOS: github.com/Incomprehensible/COSMOS
