Edge Orchestrator Planning with Distributed Ledger for SLA Compliance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge computing systems face challenges in orchestrator execution planning, particularly in selecting edge devices for workload execution that comply with service level agreements (SLAs) and key performance indicators (KPIs), which are resource-constrained and require efficient resource management and security in multi-tenant environments.
Innovation Solution
Utilizing reinforcement learning and distributed ledger techniques to generate an execution plan for workloads on edge nodes, selecting devices based on reinforcement learning reward values and distributed ledger records to achieve state transitions that meet SLAs and KPIs, while managing resource allocation and security across multiple tenants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reinforcement learning and distributed ledger techniques are used to select edge devices for workload execution, then compliance with SLAs and KPIs is improved, but system complexity increases
Solution Approach 1:
The patent introduces a distributed ledger as an intermediary layer between the orchestration system and edge devices. This ledger records and verifies device states, resource availability, and workload execution compliance, thereby improving SLA adherence without requiring direct complex interactions between all system components. The ledger acts as a trusted mediator that simplifies the orchestration logic while maintaining reliability.
Solution Approach 2:
The reinforcement learning mechanism implements continuous feedback loops where the orchestration system monitors workload execution outcomes, resource utilization, and SLA compliance metrics. This feedback is used to dynamically adjust device selection policies and resource allocation decisions, improving compliance over time. The distributed ledger also provides feedback by recording actual execution states for verification against planned states.
2Productivity
If reinforcement learning is used to optimize resource allocation across multiple tenants, then productivity is improved, but computational overhead increases
Solution Approach 1:
The system performs preliminary actions by pre-calculating and storing resource allocation policies, device capability profiles, and SLA requirements in the distributed ledger before actual workload execution. This allows the reinforcement learning algorithm to make faster decisions during runtime by querying pre-processed information rather than performing complex calculations from scratch, thereby reducing computational overhead while maintaining allocation efficiency.
3Measurement precision
If distributed ledger records are maintained for each state transition, then measurement precision is improved, but data storage requirements increase
Solution Approach 1:
The patent extracts and stores only the essential state transition information in the distributed ledger, such as device identifiers, workload identifiers, resource allocation states, and compliance metrics. Non-essential detailed data is either omitted or stored locally at edge devices. This selective extraction maintains measurement precision for critical parameters while reducing the overall data storage burden on the centralized ledger system.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and use cases for orchestrator execution planning using a distributed ledger are discussed, including an orchestration system with memory and at least one processing circuitry coupled to the memory. The processing circuitry is configured to perform operations to generate an execution plan for a workload based on an SLA. The execution plan includes state transitions associated with corresponding edge service instances. A distributed ledger record is retrieved from the ledger based on a reinforcement learning reward value specified by the record. The reward value is associated with a state transition of the plurality of state transitions. An edge node is selected based on the retrieved distributed ledger record. Execution of an edge service instance of the plurality of edge service instances by the edge node is scheduled. The execution of the edge service instance corresponds to the state transition associated with the reinforcement learning reward value.