Real World Automotive Benchmarks For Free

Real World Automotive Benchmarks For Free - A large quantity of innovative functionalities in modern automotive systems are realized using significant amounts of software technologies.

As a consequence the job of integrating many different applications onto the same target platform has grown to a more and more complex and time consuming task. One important tool for guaranteeing the correctness of the dynamic behavior of the integrated system, especially for the cyber physical parts, is timing analysis.

There exists a large body of work in the domain of worstcase timing (or real-time) analysis. Each method assumes specific application and platform models addressing a subset of existing real-world timing effects.

For cases where the application and platform models are simple enough, maximum utilization bounds can be derived to decide whether or not a given system adheres to some predefined real-time constraints.

Prominent examples for this kind of analyses are, for instance, the work of Liu and Layland for independent periodic task sets under rate-monotonic fixed priority scheduling on a single core platform, or the work of Dertouzos on earliest deadline first (EDF) scheduling.

An overview of extensions on those basic works can be found under. More complex application and platform models are addressed by approached based on the so-called busy window analysis combined with reasoning about the critical instant as proposed by Lehoczky.

As of today there exist proprietary industry strength tools based on that approach, such as SymTA/S, that are capable of analyzing most timing effects in current automotive systems. However, the detailed analysis techniques in those tools are not published, and thus, not accessible for the real-time research community.

As a result, there currently exist only few directly applicable tools and approaches that can cope with the complexity of the dynamic behavior in modern automotive systems. Therefore, simulation based methods are very popular albeit time consuming and inherently unsafe, which is unsatisfactory given the potential for front loading with formal analysis techniques.

Due to the introduction of multi-core execution platforms, the risk of divergence between academic research and industrial practice is currently increasing. The reason is the strongly increased problem space for timing analysis induced by multi-core systems.

Extending existing approaches is very challenging since the system structure and the dynamic system behavior of automotive systems is very complex. The reasons are manifold:

Control of many physical processes with strongly varying dynamics
Co-existence of sampled and reactive system parts
Different time domains (e.g. crank angle, timers, incoming network traffic) leading to complex scheduling situations
Numerous complex communication dependencies between functional entities in different time domains due to high coupling which is due to physical dependencies
Sophisticated platform mechanisms influencing the dynamic behavior (cooperative tasks for saving stack space, automated copy mechanisms for data consistency, etc.)

There exist some tools that address the generation of synthetic applications for benchmark purposes that are worth being mentioned here. For instance, the Task Graphs for Free (TGFF) tool [8] generates a set of random independent task trees.

Thereby, the structure of the generated task trees is very general, and could most likely be tweaked to fit the structure of automotive applications. This paper can help to do the mapping of the TGFF model to automotive application model including a sensible parameterization.

However, TGFF lacks a model for memory accesses that can contribute (depending on the mapping decisions) massively to the execution times especially in multi-core systems. Another tool for generating synthetic applications models is called System Models for Free (SMFF).

The SMFF tool focuses on generating models that are “ready for scheduling analysis”. For that purpose, it generates (in contrast to TGFF) not only an application graph, but also a platform graph consisting of computational and communication resources.

Additionally, a mapping of task and communication links to the platform graph along with scheduling parameters is generated. Overall, it might be hard to use SMFF to generate benchmarks representing typical automotive system.

The main reason is that important structural application and platform elements are not represented in the application and platform models. The goal of this paper is to give an insight into the structure of typical automotive real-time software systems, along with challenges for research into relevant novel analysis techniques.

To that end, the paper presents characteristics of an application with real-world complexity which is applicable to many control dominated application domains at Bosch. Based on these characteristics, real-time research groups are capable of generating expressive benchmarks for their realtime research without IP limitations.