**3.1 Tasks**

6 Will-be-set-by-IN-TECH

Faggioli et al. (2009) present an implementation of the Earliest Deadline First (EDF) and constant bandwidth servers for the Linux kernel, with support for multicore platforms. It is implemented directly into the Linux kernel. Each task is assigned a period (equal to its relative deadline) and a budget. When a task exceeds its budget, it is stopped until its next period expires and its budget is replenished. This provides temporal protection, as the task behaves like a hard reservation. Each task is assigned a timer, which is activated whenever a

Eswaran et al. (2005) describe Nano-RK, a reservation-based RTOS targeted for use in resource-constrained wireless sensor networks. It supports fixed-priority preemptive multitasking, as well as resource reservations for processor, network, sensor and energy. Only one task can be assigned to each processor reservation. Nano-RK also provides explicit support for periodic tasks, where a task can wait for its next period. Each task contains a timestamp for its next period, next replenishment and remaining budget. A one-shot timer drives the timer ISR, which (i) loops through all tasks, to update their timestamps and handle

Unlike the work presented in (Behnam et al., 2008), which implements a HSF on top of a commercial operating system, and in (Faggioli et al., 2009; Oikawa & Rajkumar, 1999; Palopoli et al., 2009), which implement reservations within Linux, our design for HSF is integrated within a RTOS targeted at embedded systems. Kim et al. (2000) describe a micro-kernel with

Our design aims at efficiency, in terms of memory and processor overheads, while minimizing the modifications of the underlying RTOS. Unlike Behnam et al. (2008); Oikawa & Rajkumar (1999); Palopoli et al. (2009) it avoids recalculating the expiration of local server events, such as budget depletion, upon every server switch. It also limits the interference of inactive servers on system level by deferring the handling of their local events until they are switched in. While Behnam et al. (2008) present an approach for limiting interference of periodic idling servers, to the best of our knowledge, our work is the first to also cover deferrable servers.

Asberg et al. (2009) make first steps towards using hierarchical scheduling in the AUTOSAR standard. They sketch what it would take to enable the integration of software components by providing temporal isolation between the AUTOSAR components. In (Nolte et al., 2009) they extend their work to systems where components share logical resources, and describe how to apply the SIRAP protocol (Behnam et al., 2007) for synchronizing access to resources shared between tasks belonging to different components. In this work we consider independent components and focus on minimizing the interference between components due to them

In this paper we assume a system is composed of independently developed and analyzed components. A components consists of a set of tasks which implement the desired application, a local scheduler, and a server. There is a one-to-one mapping between components and

task is switched in, by recalculating the deadline event for the task.

the expired events, and (ii) sets the one-shot timer to the next wakeup time.

a two-level HSF and time-triggered scheduling on the global level.

**2.3 Hierarchical scheduling in automotive systems**

sharing the timer management system.

**3. System model**

servers.

We consider a set Γ of periodic tasks, where each task *τ<sup>i</sup>* ∈ Γ is specified by a tuple (*i*, *φi*, *Ti*, *Ci*), where *i* is a fixed priority (smaller *i* means higher priority), *φ<sup>i</sup>* is the task's phasing, *Ti* is the interarrival time between two consecutive jobs, and *Ci* is its worst-case execution time. Tasks are preemptive and independent.
