February 5, 2024 | Nikos Filinis, Ioannis Tzanettis, Dimitrios Spatharakis, Eleni Fotopoulou, Ioannis Dimolitsas, Anastasios Zafeiropoulos, Constantinos Vassilakis, Symeon Papavassiliou
The paper presents an intent-driven orchestration approach for serverless applications deployed across a computing continuum, which includes edge and cloud infrastructures. The approach aims to manage serverless applications with strict Quality of Service (QoS) requirements, leveraging the distributed nature of the computing continuum to optimize resource utilization and performance. Key contributions include:
1. **Intent-driven Orchestration**: A framework that abstracts resource management from application developers and providers, allowing them to specify high-level objectives, properties, and constraints (SLOs) without detailed resource definitions. This approach supports unified management across multiple clusters and ensures optimal resource allocation.
2. **Reinforcement Learning (RL)-Driven Autoscaling**: An RL-based method for autoscaling serverless function chains, using a reward function to minimize resource usage while meeting Service Level Objectives (SLOs). An ARIMA-based workload predictor aids in timely scaling decisions.
3. **Scheduling Optimizer**: A multi-objective optimization problem formulation to decide the placement of function chains in the computing continuum, aiming to minimize link utilization, transformation costs, and power consumption. The solution minimizes inter-function communication and start-up delays while ensuring energy efficiency.
4. **Evaluation**: The proposed framework is evaluated in a small-scale testbed, demonstrating 1.52% QoS violations with a slight increase in deployed resources, significant cost reduction, and power consumption savings compared to state-of-the-art techniques.
The paper also discusses related work in intent-driven orchestration, autoscaling, and distributed function scheduling, highlighting the unique challenges and contributions of the proposed approach.The paper presents an intent-driven orchestration approach for serverless applications deployed across a computing continuum, which includes edge and cloud infrastructures. The approach aims to manage serverless applications with strict Quality of Service (QoS) requirements, leveraging the distributed nature of the computing continuum to optimize resource utilization and performance. Key contributions include:
1. **Intent-driven Orchestration**: A framework that abstracts resource management from application developers and providers, allowing them to specify high-level objectives, properties, and constraints (SLOs) without detailed resource definitions. This approach supports unified management across multiple clusters and ensures optimal resource allocation.
2. **Reinforcement Learning (RL)-Driven Autoscaling**: An RL-based method for autoscaling serverless function chains, using a reward function to minimize resource usage while meeting Service Level Objectives (SLOs). An ARIMA-based workload predictor aids in timely scaling decisions.
3. **Scheduling Optimizer**: A multi-objective optimization problem formulation to decide the placement of function chains in the computing continuum, aiming to minimize link utilization, transformation costs, and power consumption. The solution minimizes inter-function communication and start-up delays while ensuring energy efficiency.
4. **Evaluation**: The proposed framework is evaluated in a small-scale testbed, demonstrating 1.52% QoS violations with a slight increase in deployed resources, significant cost reduction, and power consumption savings compared to state-of-the-art techniques.
The paper also discusses related work in intent-driven orchestration, autoscaling, and distributed function scheduling, highlighting the unique challenges and contributions of the proposed approach.