Real-time edge framework (RTEF): task scheduling and realisation

Abstract

With the big success of the Cloud Computing, or the Cloud, new research areas appeared. Edge Computing (EC) is one of the recent paradigms that is expected to overcome the Quality of Service (QoS) and latency issues caused by the best-effort behaviour of the Cloud. EC aims to bring the computation power close to the end devices as much as possible and reduce the dependency to the Cloud. Bringing computing power close to the source also enables real-time applications. In this paper, we propose a novel software reference architecture for Edge Servers, which is operating system (OS) and hardware-agnostic. Edge Servers can collaborate and execute (near) real-time tasks on time, either by downscaling or scheduling them according to their deadlines or offloading them to other Edge Servers in the network. Decision making for offloading, resource planning, and task scheduling are challenging problems in decentralized systems. The paper explains how resource planning and task scheduling can be overcome with software approach. Finally, the article realises the architecture as a framework, called Real-Time Edge Framework (RTEF) and validates its correctness with a use case.

Real-Time Edge Framework (RTEF): Decentralized Decision Making for Offloading

“…A service has several parameters to be set, based on the software behaviour such as execution duration, relative deadline, CPU affinities, and execution utilization (maximum CPU utilization/bandwidth). Based on the three task types, we introduced three service types that wrap tasks depending on their behaviours [7]:…”

“…Once the task is received, first, the server lists all available Edge Servers that have the service defined for that task. If the receiving server has enough resources to run the task, it plans the execution [7]. If not, an Edge Server in the Edge Network is searched, which can complete the execution in the shortest time.…”

“…If the sum of execution utilizations (CPU utilization) of all running tasks, including the requested task, does not exceed 100% at each CPU, then the tasks can run without further action. Otherwise, one possibility is to scale down the CPU utilization of the newly requested task, already running tasks or all tasks [7].…”

“…For tasks in Legacy and Simple type, Non-resumable And Preemptible Aperiodic TAsk (NAPATA) scheduling is used. The novel NAPATA scheduling algorithm is explained in our another paper [7]. If the scheduling is also not possible, the task fails to execute.…”

“…Otherwise, the Edge Server must choose an optimal Edge Server to offload the task to. The task scheduling methods to plan the execution is explained in our other work [7]. This paper will explain how an Edge Server makes decisions, either to offload or execute a task on itself.…”